From f80d26e5aff12f79d2a2addc5dcfa622a316ccbc Mon Sep 17 00:00:00 2001
From: Chris Phillipson <christopher.a.phillipson@gmail.com>
Date: Wed, 23 Sep 2026 09:29:58 -0700
Subject: [PATCH] feat(models): Opus 5.5 and GPT-6 Sol/Luna pricing and
 current-generation defaults

Pricing (verified 2026-09-23 against first-party pages):
- Add claude-opus-5-5 ($4/$20, cache reads 0.05x), gpt-6-sol ($2/$10) and
  gpt-6-luna ($0.10/$0.50), all released 2026-09-22.
- claude-opus-5-5 previously matched claude-opus-5 on the token boundary and
  was costed at $5/$25 with 0.1x cache reads; real local usage was overstated
  by 94%.
- Provider-namespaced ids (anthropic/claude-opus-4.7) now match on their last
  path segment instead of falling to the fallback rate.
- codex-auto-review stays on the flagged fallback: OpenAI publishes no rate.

Defaults:
- Claude reasoning routes and the implementation/testing escalation rung move
  to claude-opus-5-5, Anthropic's recommended starting model.
- Codex execution moves to gpt-6-sol and mechanical work to gpt-6-luna. Codex's
  catalog ranks GPT-6 first and labels GPT-5.6 "Older".
- Cross-host tiers pair by role and price: Sol <-> Sonnet 5, Astra <-> Fable 5.1,
  Luna <-> Haiku. Earlier models stay listed for pins; seeded routes report
  divergence until `ak x host refresh`.

Catalog: Opus 5.5 lifecycle facts; OpenAI doc links for Sol and Luna.
Tests derive their "prior catalog" fixtures from DEFAULT_ROUTES.
Docs: MODEL-PRICING-AUDIT, USAGE-SCORECARD-METRICS, PROVIDERS, MODELS, ADR-0002.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 docs/MODEL-PRICING-AUDIT.md                   | 155 +++++++++++-------
 docs/MODELS.md                                |   4 +-
 docs/PROVIDERS.md                             |  13 +-
 docs/USAGE-SCORECARD-METRICS.md               |  63 +++----
 ...-vocabulary-defaults-from-ruv-templates.md |  11 +-
 src/commands/models.mjs                       |   4 +-
 src/commands/x/host.mjs                       |   2 +-
 .../discovery/anthropic-catalog.mjs           |   5 +-
 src/lib/model-inventory/read-model.mjs        |   2 +
 src/lib/pricing.mjs                           |  39 +++--
 src/lib/routing.mjs                           |  44 +++--
 .../dashboard-integration-identity.test.mjs   |   3 +-
 tests/kit/model-dashboard-read-model.test.mjs |   2 +-
 .../kit/model-discovery-claude-codex.test.mjs |  13 +-
 tests/kit/pricing.test.mjs                    |  43 ++++-
 tests/kit/routing-divergence.test.mjs         |  16 +-
 tests/kit/status-viability.test.mjs           |   2 +-
 17 files changed, 282 insertions(+), 139 deletions(-)

diff --git a/docs/MODEL-PRICING-AUDIT.md b/docs/MODEL-PRICING-AUDIT.md
index 8fcf340e..b5c09f97 100644
--- a/docs/MODEL-PRICING-AUDIT.md
+++ b/docs/MODEL-PRICING-AUDIT.md
@@ -1,66 +1,107 @@
-# Model listing and pricing audit — 2026-09-08
+# Model listing and pricing audit — 2026-09-23
 
 The maintained OpenAI and Anthropic text-token rate entries were checked against
 provider-owned public documentation. API-equivalent estimates are not subscription
-bills or evidence of account access. Existing routing defaults remain unchanged.
+bills or evidence of account access. Routing defaults moved to the current generation (see below).
+
+## Additions
+
+- **Claude Opus 5.5** (`claude-opus-5-5`, released 2026-09-22): $4 input, $5
+  five-minute writes, $8 one-hour writes, $0.20 cache reads and $20 output per
+  million tokens. Cache reads are 0.05× input, not the usual 0.1×. 1M context,
+  128K output, retirement not sooner than 2027-09-22.
+  [Pricing](https://platform.claude.com/docs/en/about-claude/pricing),
+  [models overview](https://platform.claude.com/docs/en/about-claude/models/overview),
+  [deprecations](https://platform.claude.com/docs/en/about-claude/model-deprecations).
+- **GPT-6 Sol** (`gpt-6-sol`): $2 input, $0.20 cached input, $2.50 cache writes,
+  $10 output. [Model page](https://developers.openai.com/api/docs/models/gpt-6-sol).
+- **GPT-6 Luna** (`gpt-6-luna`): $0.10 input, $0.01 cached input, $0.125 cache
+  writes, $0.50 output. [Model page](https://developers.openai.com/api/docs/models/gpt-6-luna).
+
+Both GPT-6 models were announced on 2026-09-22
+([SiliconANGLE](https://siliconangle.com/2026/09/22/anthropic-releases-claude-opus-5-5-and-openai-counters-with-two-cheaper-gpt-6-models/)).
+Their model pages list April 20 and May 18, 2026 as the "release date", which
+match the stated knowledge cutoffs, not the public launch. Both have a 1.05M
+context window, 128K output and the same >272K request surcharge as Astra,
+which the estimator does not model.
+
+## Routing defaults
+
+- Claude reasoning activities (architecture, design, security-analysis,
+  debugging, and the implementation/testing escalation rung) now default to
+  `claude-opus-5-5`, Anthropic's recommended starting model
+  ([models overview](https://platform.claude.com/docs/en/about-claude/models/overview)).
+  Opus 5 remains listed for pins.
+- Codex implementation, testing and security-scan now default to `gpt-6-sol`,
+  documentation and packaging to `gpt-6-luna`. Codex's own model catalog ranks the
+  GPT-6 models first and labels GPT-5.6 "Older"; OpenAI's Codex docs make Sol the
+  default preset ([Codex models](https://learn.chatgpt.com/docs/models)). GPT-5.6
+  models remain listed for pins.
+- Cross-host tiers pair by role and per-token price: GPT-6 Sol ↔ Sonnet 5
+  ($2/$10), GPT-6 Astra ↔ Fable 5.1 ($10/$50), GPT-6 Luna ↔ Haiku 4.5.
+- Routes seeded before this change are reported as diverged and keep their model
+  until `ak x host refresh`.
+- Not added as retirements: OpenAI's Codex docs retire GPT-5.4 / 5.4 Mini
+  (2026-08-31) and GPT-5.5 (2026-10-14) for ChatGPT sign-in, but the API still
+  serves them, so a read-time substitution would override API-key users' pins.
 
 ## Corrections
 
-- Added GPT-6 Astra at $10 input, $1 cached input, $12.50 cache writes and $50 output
-  per million tokens. [OpenAI model page](https://developers.openai.com/api/docs/models/gpt-6-astra).
-- Confirmed Fable 5.1 and Mythos 5.1 already exist with $10 input, $0.25 cache reads,
-  $12.50 five-minute writes and $50 output. One-hour writes cost $20 and remain
-  outside the estimator's five-minute assumption. [Anthropic pricing](https://platform.claude.com/docs/en/about-claude/pricing).
-- Removed Sonnet 5's canceled September increase: $2/$10 is now standard.
-  [Anthropic pricing notice](https://platform.claude.com/docs/en/about-claude/pricing).
-- Resolved the exact `gpt-5.6` alias to Sol; unknown variants remain unmatched.
-  [Sol model page](https://developers.openai.com/api/docs/models/gpt-5.6-sol).
-- Removed the unsupported cached-input discount on Pro models. GPT-5.5 Pro
-  explicitly has no discount; GPT-5.4 Pro publishes no cached rate, so the
-  estimator conservatively uses ordinary input rates for reported cached tokens.
-  [GPT-5.5 Pro](https://developers.openai.com/api/docs/models/gpt-5.5-pro),
-  [GPT-5.4 Pro](https://developers.openai.com/api/docs/models/gpt-5.4-pro).
-- Applied OpenAI's 1.25× write premium only to GPT-5.6 and later; older OpenAI
-  entries use ordinary input rates. [Prompt caching](https://developers.openai.com/api/docs/guides/prompt-caching).
-- Removed GLM 5's static “cheapest” claim. Current OpenRouter listings put
-  GLM 5.2 below GLM 5 in per-token price; provider offers can change.
-  [GLM 5.2](https://openrouter.ai/z-ai/glm-5.2), [GLM 5](https://openrouter.ai/z-ai/glm-5).
-
-## Model facts and retained rates
-
-Fable 5.1 and Mythos 5.1 have 1M context and 128K maximum output. Mythos is
-invitation-only; listing it as a choice does not assert entitlement.
-[Official overview](https://platform.claude.com/docs/en/models/fable-5-1/overview).
-The remaining bundled Anthropic lifecycle entries match the
-[deprecation history](https://platform.claude.com/docs/en/about-claude/model-deprecations);
-Mythos availability is established by the model overview, not the retirement table.
-
-OpenAI base input/output rates below were verified and retained (USD/MTok):
-
-| Model | Input | Output | Source |
-|---|---|---|---|
-| GPT-5.6 Sol | 4 | 20 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-| GPT-5.6 Terra | 2 | 12 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-| GPT-5.6 Luna | 0.20 | 1.20 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-| GPT-5.5 | 5 | 30 | [Model](https://developers.openai.com/api/docs/models/gpt-5.5) |
-| GPT-5.5 Pro | 30 | 180 | [Model](https://developers.openai.com/api/docs/models/gpt-5.5-pro) |
-| GPT-5.4 | 2.50 | 15 | [Model](https://developers.openai.com/api/docs/models/gpt-5.4) |
-| GPT-5.4 Mini | 0.75 | 4.50 | [Model](https://developers.openai.com/api/docs/models/gpt-5.4-mini) |
-| GPT-5.4 Nano | 0.20 | 1.25 | [Model](https://developers.openai.com/api/docs/models/gpt-5.4-nano) |
-| GPT-5.4 Pro | 30 | 180 | [Model](https://developers.openai.com/api/docs/models/gpt-5.4-pro) |
-| GPT-5.3 Codex | 1.75 | 14 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-| chat-latest | 5 | 30 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-| Realtime 2.1 (text) | 4 | 24 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-| Realtime 2.1 Mini (text) | 0.60 | 2.40 | [Pricing](https://developers.openai.com/api/docs/pricing) |
-
-Sol's promotional price is guaranteed at least through November 21, 2026;
-no definite expiry is published. No future price is invented. Retained historical
-rates do not imply that a host still offers a model.
+- `claude-opus-5-5` had no entry and matched `claude-opus-5` on the token
+  boundary, so real Opus 5.5 usage was costed at $5/$25 with 0.1× cache reads.
+  It now resolves to its own row.
+- Provider-namespaced ids (`anthropic/claude-opus-4.7`, `openai/gpt-5.5`) fell
+  through to the fallback rate because matching is by prefix. They now match on
+  their last path segment.
+- The previous audit said one-hour cache writes were outside the estimator.
+  They are priced at 2× input from the transcript's TTL split.
+
+## Verified and unchanged
+
+All other Anthropic entries match the live pricing table, including Fable 5.1 /
+Mythos 5.1 (0.025× cache reads) and Sonnet 5 at $2/$10. No Anthropic model was
+deprecated since the last audit; Sonnet 4.5 and Haiku 4.5 remain Active.
+
+OpenAI base rates (USD/MTok), checked against
+[OpenAI pricing](https://developers.openai.com/api/docs/pricing):
+
+| Model | Input | Output |
+|---|---|---|
+| GPT-6 Astra | 10 | 50 |
+| GPT-5.6 Sol | 4 | 20 |
+| GPT-5.6 Terra | 2 | 12 |
+| GPT-5.6 Luna | 0.20 | 1.20 |
+| GPT-5.5 | 5 | 30 |
+| GPT-5.5 Pro | 30 | 180 |
+| GPT-5.4 | 2.50 | 15 |
+| GPT-5.4 Mini | 0.75 | 4.50 |
+| GPT-5.4 Nano | 0.20 | 1.25 |
+| GPT-5.4 Pro | 30 | 180 |
+| GPT-5.3 Codex | 1.75 | 14 |
+| chat-latest | 5 | 30 |
+| Realtime 2.1 (text) | 4 | 24 |
+| Realtime 2.1 Mini (text) | 0.60 | 2.40 |
+
+GPT-5.6 Sol's promotional price is guaranteed at least through November 21,
+2026; no future price is invented. Some third-party summaries list GPT-5.6 Sol
+at $5/$30; OpenAI's own pricing page shows $4/$20.
+
+## Not priced, deliberately
+
+- `codex-auto-review`, Codex's automatic approval reviewer, appears in local
+  Codex rollouts with real token counts. OpenAI publishes no per-token price for
+  it. It is a server-side alias: Codex
+  [PR #18169](https://github.com/openai/codex/pull/18169) replaced a hardcoded
+  `gpt-5.4` with it so the backend can choose the model. It stays unmatched and
+  is costed at the flagged fallback rate rather than at a guessed one.
+- `gpt-reserve` (Luna Reserve) is a hidden Codex fallback allowance with no
+  published API price.
 
 ## Estimator limits
 
-Request-level long-context surcharges, service modes, regional uplifts, audio/image
-rates and one-hour cache writes are not inferred from daily aggregate token counts.
-OpenAI's API cache-write counter is included in total input; native host transcripts
-must expose equivalent evidence before their parsers can attribute those writes.
-The model list and pricing table do not constitute a complete provider catalog.
+Request-level long-context surcharges, service modes (Batch, Flex, Priority,
+Fast), regional uplifts and audio/image rates are not inferred from daily
+aggregate token counts. OpenAI's API cache-write counter is included in total
+input; native host transcripts must expose equivalent evidence before their
+parsers can attribute those writes. The model list and pricing table do not
+constitute a complete provider catalog.
diff --git a/docs/MODELS.md b/docs/MODELS.md
index 8fdc36a9..ca723970 100644
--- a/docs/MODELS.md
+++ b/docs/MODELS.md
@@ -17,8 +17,8 @@ ak models refresh
 ak models refresh --host claude
 ak models status
 ak models diff
-ak models explain codex:gpt-5.6-terra
-ak models plan --activity testing --from codex:gpt-5.4 --to codex:gpt-5.6-terra
+ak models explain codex:gpt-6-sol
+ak models plan --activity testing --from codex:gpt-5.6-terra --to codex:gpt-6-sol
 ```
 
 Only `refresh` contacts model sources or writes a snapshot. `status`, `diff`, `explain`, `plan`,
diff --git a/docs/PROVIDERS.md b/docs/PROVIDERS.md
index 9ad1a44b..bf0adaf0 100644
--- a/docs/PROVIDERS.md
+++ b/docs/PROVIDERS.md
@@ -373,7 +373,7 @@ When you want explicit ordering rather than env auto-enable, `ak` manages agenti
 ```bash
 ak host pick \
   --aqe-provider claude-code \
-  --aqe-fallback 'claude-code:claude-opus-5; openai:gpt-5.6; gemini:gemini-3.5-flash'
+  --aqe-fallback 'claude-code:claude-opus-5-5; openai:gpt-5.6; gemini:gemini-3.5-flash'
 ```
 
 Each `provider:model,model` becomes an ordered chain entry (first = highest priority). `ak`
@@ -434,7 +434,7 @@ provider — add them to the chain and put `OPENROUTER_API_KEY` in your env:
 ```bash
 ak host pick \
   --aqe-provider claude-code \
-  --aqe-fallback 'claude-code:claude-opus-5; openrouter:z-ai/glm-5.2'
+  --aqe-fallback 'claude-code:claude-opus-5-5; openrouter:z-ai/glm-5.2'
 ```
 
 Curated picks (verified September 8, 2026): `z-ai/glm-5.2` (flagship — 1M context, strong
@@ -507,12 +507,15 @@ Defaults (all overridable; your edits are marked `custom` and never re-seeded):
 | Activity | Host | Default model |
 |---|---|---|
 | specification, review, release | claude | `claude-sonnet-5` |
-| architecture, design, debugging, security-analysis | claude | `claude-opus-5` |
-| implementation, testing, security-scan | codex | `gpt-5.6-terra` |
-| documentation, packaging | codex | `gpt-5.6-luna` |
+| architecture, design, debugging, security-analysis | claude | `claude-opus-5-5` |
+| implementation, testing, security-scan | codex | `gpt-6-sol` |
+| documentation, packaging | codex | `gpt-6-luna` |
 
 *(packaging & release are `ak`-added — ruflo ships templates for feature/security/refactor only.)*
 
+Implementation and testing escalate to `claude-opus-5-5`. Routes seeded before a default
+changes are reported as diverged and keep their model until you run `ak x host refresh`.
+
 **Retired Codex models.** `ak` has no automatic Codex retirement substitutions as of 2026-08-25.
 The current [OpenAI API model catalog](https://developers.openai.com/api/docs/models/all) still lists
 GPT-5.4 and GPT-5.4 mini, and no first-party withdrawal notice supports the former automatic
diff --git a/docs/USAGE-SCORECARD-METRICS.md b/docs/USAGE-SCORECARD-METRICS.md
index 3baed825..210f67ae 100644
--- a/docs/USAGE-SCORECARD-METRICS.md
+++ b/docs/USAGE-SCORECARD-METRICS.md
@@ -515,7 +515,7 @@ cost = (inputUnits × rate_in + output × rate_out) / 1,000,000
 Mythos 5.1, which resolve to 0.025, and OpenAI Pro, which uses 1 (§13). Summed across every row in the
 window.
 
-**Source:** `costOf()`, `src/lib/pricing.mjs:304-315`, reproduced verbatim:
+**Source:** `costOf()`, `src/lib/pricing.mjs`, reproduced verbatim:
 
 ```js
 export function costOf(usage) {
@@ -567,12 +567,12 @@ so the estimator conservatively uses ordinary input rates for reported cached to
 See [OpenAI prompt caching](https://developers.openai.com/api/docs/guides/prompt-caching).
 
 Rate resolution is **longest-prefix match** (`isPrefixOf`, `KEYS_BY_LENGTH`)
-on a normalized model id (`pricing.mjs:205-214`), so a dated release
+on a normalized model id (`pricing.mjs`), so a dated release
 (`claude-haiku-4-5-20251001`)
 resolves to the same entry as its bare alias, and a more specific entry
 (`gpt-5.6-sol`) is never shadowed by a shorter one (`gpt-5.6`). An id matching
 nothing gets `FALLBACK_PRICE` — Sonnet-class rate, `$3`/`$15`
-(`pricing.mjs:182`) — rather than `$0`, so an unrecognized model can never be
+(`pricing.mjs`) — rather than `$0`, so an unrecognized model can never be
 silently free; `matched: false` travels with the result so a maintainer can
 find fallback-priced rows if the table needs a new entry.
 
@@ -582,7 +582,7 @@ Each table entry is a **schedule** — an ordered list of periods, each with the
 day it takes effect. Nearly every entry has exactly one period that has always
 applied (`anthropic(5, 25)` builds that shape); a confirmed effective rate
 change can be represented by multiple periods (`schedule`,
-`pricing.mjs:45-68`). `periodOn` (`pricing.mjs:229`) picks the last period
+`pricing.mjs`). `periodOn` (`pricing.mjs`) picks the last period
 already in effect on the given day, comparing ISO date strings
 lexicographically so no `Date` parsing is involved and the module stays
 clock-free.
@@ -608,7 +608,7 @@ Two rules bound the mechanism:
   Codex promo would be a one-line edit rather than new machinery. Rates that
   vary by *how* a request was served — regional uplift, large-prompt surcharge,
   service tiers — are a different axis, are deliberately **not** expressible
-  here, and remain in `UNMODELLED_PRICING_FACTORS` (`pricing.mjs:182-185`)
+  here, and remain in `UNMODELLED_PRICING_FACTORS` (`pricing.mjs`)
   because a transcript does not record the endpoint or tier.
 
 A `priceFor` call with no day prices as of `PRICES_AS_OF`, the table's
@@ -721,7 +721,7 @@ rates; the subtitle is overgeneralized and needs a UI correction.
 **Why this number matters more than it looks like it should.** On the
 reference corpus, 96.3% of tokens were cache reads — pricing them as fresh
 input (rather than at the 0.1× multiplier) would overstate cost by roughly
-10× (`pricing.mjs:7-10`). A cache-read share this high is not an anomaly to be
+10× (`pricing.mjs`). A cache-read share this high is not an anomaly to be
 suspicious of on its own — it is the expected steady state for any
 long-running agentic session that resends a large, mostly-unchanged system
 prompt and tool-result history on every turn, which both Claude Code and
@@ -1154,13 +1154,13 @@ public documentation; they are maintained data, not live quotes.
 ### 13.1 Anthropic — primary source, directly verified
 
 Rates rechecked against [Anthropic's official pricing](https://platform.claude.com/docs/en/about-claude/pricing)
-on 2026-09-08. Fable 5.1 and Mythos 5.1 were already catalogued with the
-correct rates; this pass refreshes their verification evidence.
+on 2026-09-23, which adds Claude Opus 5.5 (released 2026-09-22).
 
 | Model | Base input | 5m cache write | 1h cache write | Cache read (hit) | Output |
 |---|---|---|---|---|---|
 | Claude Fable 5.1 / Mythos 5.1 | $10/MTok | $12.50/MTok | $20/MTok | $0.25/MTok¹ | $50/MTok |
 | Claude Fable 5 / Mythos 5 | $10/MTok | $12.50/MTok | $20/MTok | $1/MTok | $50/MTok |
+| Claude Opus 5.5 | $4/MTok | $5/MTok | $8/MTok | $0.20/MTok² | $20/MTok |
 | Claude Opus 5 | $5/MTok | $6.25/MTok | $10/MTok | $0.50/MTok | $25/MTok |
 | Claude Opus 4.8 / 4.7 / 4.6 / 4.5 | $5/MTok | $6.25/MTok | $10/MTok | $0.50/MTok | $25/MTok |
 | Claude Sonnet 5 | $2/MTok | $2.50/MTok | $4/MTok | $0.20/MTok | $10/MTok |
@@ -1168,8 +1168,8 @@ correct rates; this pass refreshes their verification evidence.
 | Claude Haiku 4.5 | $1/MTok | $1.25/MTok | $2/MTok | $0.10/MTok | $5/MTok |
 
 ¹ Cache hits on Claude Fable 5.1 and Claude Mythos 5.1 price at **0.025×** base
-input — every other current Anthropic model uses 0.1×. Verified 2026-09-08
-against **[C1]**'s live pricing page.
+input. ² Cache hits on Claude Opus 5.5 price at **0.05×**. Every other current
+Anthropic model uses 0.1×. Verified 2026-09-23 against **[C1]**'s live pricing page.
 
 Sonnet 5's previously announced September increase was canceled by Anthropic;
 $2/$10 is now its standard price. Fable 5.1 is generally available; Mythos 5.1
@@ -1182,15 +1182,15 @@ cache-read *columns* in this table are provider-published absolute rates; the
 kit's `pricing.mjs` instead stores **multipliers** — 1.25× for a 5-minute cache
 write and 2× for a 1-hour cache write (both uniform, no published per-model
 exception) and, for cache reads, 0.1×
-for every model *except* Fable 5.1 / Mythos 5.1, which carry their own
-0.025× cache-read override on that catalog entry (`pricing.mjs:90-91`)
+for every model *except* Fable 5.1 / Mythos 5.1 (0.025×) and Opus 5.5 (0.05×),
+which carry their own cache-read override on their `PRICES` entries
 instead of the module-wide default multiplier.
 
-`priceFor()` (`pricing.mjs:261-280`) is what resolves that: it returns a
+`priceFor()` (`pricing.mjs`) is what resolves that: it returns a
 `cacheReadMultiplier` field taken from the matched entry, falling back to
-`CACHE_READ_MULTIPLIER` (`pricing.mjs:186-188`) when the entry carries none.
+`CACHE_READ_MULTIPLIER` (`pricing.mjs`) when the entry carries none.
 
-`costOf()` (`pricing.mjs:304-315`) then multiplies cache-read tokens by
+`costOf()` (`pricing.mjs`) then multiplies cache-read tokens by
 that resolved value rather than a hardcoded constant.
 
 Claude transcripts record which TTL each cache write used
@@ -1201,7 +1201,7 @@ and `costOf` prices it at 2×. A record with no split prices every write at the
 rate is exactly *$5 × 1.25*, its $10 1-hour rate exactly *$5 × 2*; its published
 $0.50 cache-read rate is exactly *$5 × 0.1*. Every row in Anthropic's own table
 satisfies `cache_write_5m = input × 1.25`, `cache_write_1h = input × 2` and, except for the Fable 5.1 / Mythos 5.1
-row noted above, `cache_read = input × 0.1` — confirming the multiplier
+and Opus 5.5 rows noted above, `cache_read = input × 0.1` — confirming the multiplier
 approach is arithmetically identical to using the provider's published
 absolute cache rates directly.
 
@@ -1210,11 +1210,11 @@ in prose, independent of the pricing table: *"5-minute cache write tokens are
 1.25 times the base input tokens price... 1-hour cache write tokens are 2 times
 the base input tokens price... Cache read tokens are 0.1 times the
 base input tokens price."* This is the second, independent confirmation of
-`CACHE_READ_MULTIPLIER`/`CACHE_WRITE_MULTIPLIER`/`CACHE_WRITE_1H_MULTIPLIER` (`pricing.mjs:185-188`).
+`CACHE_READ_MULTIPLIER`/`CACHE_WRITE_MULTIPLIER`/`CACHE_WRITE_1H_MULTIPLIER` (`pricing.mjs`).
 
 ### 13.2 OpenAI (Codex) — hand-maintained, no canonical machine-readable source
 
-`pricing.mjs`'s own comment (`pricing.mjs:116-119`) records that
+`pricing.mjs`'s own comment (`pricing.mjs`) records that
 `~/.codex/models_cache.json` was checked directly and contains **zero**
 price-related keys — Codex CLI does not ship pricing data locally, unlike
 Anthropic which publishes a fetchable pricing document. OpenAI's rates in
@@ -1223,37 +1223,44 @@ documentation and are the most drift-prone entries in the file — this is
 explicitly why `PRICES_AS_OF` is surfaced in the UI (`u-asof`,
 `dashboard/client.mjs`) rather than assumed current.
 
-Standard USD rates per million tokens (Astra verified 2026-09-08 against
-[OpenAI's model documentation](https://developers.openai.com/api/docs/models/gpt-6-astra);
-GPT-5.6 rates reconciled with the implementation and
-[OpenAI pricing](https://developers.openai.com/api/docs/pricing)):
+Standard USD rates per million tokens, verified 2026-09-23 against
+[OpenAI pricing](https://developers.openai.com/api/docs/pricing) and the
+individual model pages for [Astra](https://developers.openai.com/api/docs/models/gpt-6-astra),
+[GPT-6 Sol](https://developers.openai.com/api/docs/models/gpt-6-sol) and
+[GPT-6 Luna](https://developers.openai.com/api/docs/models/gpt-6-luna):
 
 | Model (kit key) | Input | Output | Cache read |
 |---|---|---|---|
 | `gpt-6-astra` | $10 | $50 | $1 |
+| `gpt-6-sol` | $2 | $10 | $0.20 |
+| `gpt-6-luna` | $0.10 | $0.50 | $0.01 |
 | `gpt-5.6-sol` | $4 | $20 | $0.40 |
 | `gpt-5.6-terra` | $2 | $12 | $0.20 |
 | `gpt-5.6-luna` | $0.20 | $1.20 | $0.02 |
 | `gpt-5.5` | $5 | $30 | $0.50 |
 | `gpt-5.5-pro` | $30 | $180 | Not published |
 
-Astra cache writes cost $12.50/MTok (1.25× input), using the existing
-cache-write arithmetic. Sol's promotional rate has no confirmed end date:
+GPT-6 and GPT-5.6 cache writes cost 1.25× input (Astra $12.50, GPT-6 Sol
+$2.50, GPT-6 Luna $0.125/MTok), using the existing cache-write arithmetic.
+Do not confuse GPT-6 Sol ($2/$10) with GPT-5.6 Sol ($4/$20); both are current. Sol's promotional rate has no confirmed end date:
 OpenAI says at least through 2026-11-21, so no future reversion is invented.
 These are API list-price equivalents, not subscription charges or access guarantees.
 
-The full maintained rate table was rechecked on 2026-09-08; `PRICES_AS_OF`
+The full maintained rate table was rechecked on 2026-09-23; `PRICES_AS_OF`
 and the Usage summary now show that date. Individual entries can still override
 verification dates. The official `gpt-5.6` alias resolves exactly to Sol;
-unknown suffixed variants do not inherit its price. Realtime entries describe
+unknown suffixed variants do not inherit its price. A provider-namespaced id
+(`anthropic/claude-opus-4.7`) is matched on its last path segment. Codex's
+`codex-auto-review` (its auto-approval reviewer) has no published per-token
+price, so it stays unmatched and is costed at the flagged fallback rate. Realtime entries describe
 text tokens only; audio and image charges are outside this estimator.
 
 See [the pricing audit](MODEL-PRICING-AUDIT.md) for source links and scope.
 
 ### 13.3 What the pricing table deliberately does not model
 
-Recorded verbatim from `pricing.mjs:153-181` (`UNMODELLED_PRICING_FACTORS`,
-`pricing.mjs:182-185`) because listing known gaps is what makes the
+Recorded verbatim from `pricing.mjs` (`UNMODELLED_PRICING_FACTORS`,
+`pricing.mjs`) because listing known gaps is what makes the
 *modelled* factors credible:
 
 - **Regional-processing uplift.** OpenAI charges +10% on data-residency
diff --git a/docs/adr/0002-activity-vocabulary-defaults-from-ruv-templates.md b/docs/adr/0002-activity-vocabulary-defaults-from-ruv-templates.md
index db9251e2..82c149d3 100644
--- a/docs/adr/0002-activity-vocabulary-defaults-from-ruv-templates.md
+++ b/docs/adr/0002-activity-vocabulary-defaults-from-ruv-templates.md
@@ -2,7 +2,8 @@
 
 - **Status:** Amended by [ADR-0020](0020-ga-stable-surfaces.md)
 - **Date:** 2026-07-23
-- **Updated:** 2026-09-09 — reconciled against repository source and tests for issue #211
+- **Updated:** 2026-09-23 — default models moved to the current generation: `claude-opus-5-5`, `gpt-6-sol`, `gpt-6-luna`
+- **Previous update:** 2026-09-09 — reconciled against repository source and tests for issue #211
 - **Earlier update:** 2026-07-30
 - **Update note:** Preserved the activity vocabulary and moved its runtime ownership to the
   in-repository host-neutral runner.
@@ -49,9 +50,11 @@ model for execution) and are treated as **soft defaults** — see the "open ques
 
 Tier is the pairing key, not the model id: `MODEL_CATALOG` spells `flagship`/`balanced`/`fast`
 identically on both hosts so primary-host mirroring can map a route to its counterpart's equivalent.
-As of 2026-08-07 execution routes to `gpt-5.6-terra` (balanced) and mechanical work to `gpt-5.6-luna`
-(fast), as agentic-kit default preferences, not as evidence that `gpt-5.4` or
-`gpt-5.4-mini` were withdrawn; deep reasoning routes to `claude-opus-5`. Withdrawn ids are handled by the retirement
+As of 2026-09-23 execution routes to `gpt-6-sol` (balanced) and mechanical work to `gpt-6-luna`
+(fast), and deep reasoning routes to `claude-opus-5-5`. These are agentic-kit default preferences,
+following each vendor's recommended starting model, not evidence that earlier models were withdrawn.
+Cross-host tiers pair by role and per-token price: `gpt-6-sol` ↔ `claude-sonnet-5`,
+`gpt-6-astra` ↔ `claude-fable-5-1`, `gpt-6-luna` ↔ Haiku. Withdrawn ids are handled by the retirement
 mechanism in [ADR-0003](0003-auto-seed-dual-host-provenance.md), not by editing this table alone.
 
 ## Consequences
diff --git a/src/commands/models.mjs b/src/commands/models.mjs
index 25369a25..89026435 100644
--- a/src/commands/models.mjs
+++ b/src/commands/models.mjs
@@ -40,8 +40,8 @@ Examples:
   ak models refresh --all
   ak models status --host codex --json
   ak models diff models:before models:after
-  ak models explain codex:gpt-5.6-terra
-  ak models plan --activity testing --to codex:gpt-5.6-terra`;
+  ak models explain codex:gpt-6-sol
+  ak models plan --activity testing --to codex:gpt-6-sol`;
 
 const ALL_OWNERS = Object.freeze(['claude', 'codex', 'opencode', 'ollama']);
 
diff --git a/src/commands/x/host.mjs b/src/commands/x/host.mjs
index b0a50f26..dfe3f464 100644
--- a/src/commands/x/host.mjs
+++ b/src/commands/x/host.mjs
@@ -127,7 +127,7 @@ Options (pick, all optional — omit for interactive):
                                  needs OPENROUTER_API_KEY in the env)
   --provider <csv>             register ruflo providers (e.g. ollama:qwen3.6:27b)
   --route 'act:host[:model]'   override one activity's routing (repeatable), e.g.
-                                 --route 'implementation:claude:claude-opus-5'
+                                 --route 'implementation:claude:claude-opus-5-5'
                                  activities: specification, architecture, design,
                                  implementation, testing, review, security-scan,
                                  security-analysis, documentation, debugging,
diff --git a/src/lib/model-inventory/discovery/anthropic-catalog.mjs b/src/lib/model-inventory/discovery/anthropic-catalog.mjs
index ffdde46b..d9d60c18 100644
--- a/src/lib/model-inventory/discovery/anthropic-catalog.mjs
+++ b/src/lib/model-inventory/discovery/anthropic-catalog.mjs
@@ -1,7 +1,7 @@
 import { PRICES_AS_OF, priceFor } from '../../pricing.mjs';
 import { modelRecord, sourceRecord } from './index.mjs';
 
-export const ANTHROPIC_PUBLIC_CATALOG_AS_OF = '2026-09-02';
+export const ANTHROPIC_PUBLIC_CATALOG_AS_OF = '2026-09-23';
 export const ANTHROPIC_MODELS_URL =
   'https://platform.claude.com/docs/en/about-claude/models/overview';
 export const ANTHROPIC_LIFECYCLE_URL =
@@ -67,6 +67,9 @@ export const ANTHROPIC_PUBLIC_MODELS = Object.freeze([
     availability: 'limited', replacement: 'claude-mythos-5',
     contextLimit: 1_000_000, outputLimit: 128_000,
   },
+  active('claude-opus-5-5', 'Claude Opus 5.5', {
+    contextLimit: 1_000_000, outputLimit: 128_000, retirementNotBefore: '2027-09-22',
+  }),
   active('claude-opus-5', 'Claude Opus 5', {
     contextLimit: 1_000_000, outputLimit: 128_000, retirementNotBefore: '2027-07-24',
   }),
diff --git a/src/lib/model-inventory/read-model.mjs b/src/lib/model-inventory/read-model.mjs
index 50c61a30..3ea7d338 100644
--- a/src/lib/model-inventory/read-model.mjs
+++ b/src/lib/model-inventory/read-model.mjs
@@ -182,6 +182,8 @@ const OFFICIAL_CLAUDE_IDS = new Set(ANTHROPIC_OFFICIAL_MODEL_IDS);
 const OPENAI_MODEL_DOCUMENTATION = new Map([
   ['gpt-5.6', 'https://developers.openai.com/api/docs/models/gpt-5.6-sol'],
   ['gpt-6-astra', 'https://developers.openai.com/api/docs/models/gpt-6-astra'],
+  ['gpt-6-sol', 'https://developers.openai.com/api/docs/models/gpt-6-sol'],
+  ['gpt-6-luna', 'https://developers.openai.com/api/docs/models/gpt-6-luna'],
   ['gpt-5.6-sol', 'https://developers.openai.com/api/docs/models/gpt-5.6-sol'],
   ['gpt-5.6-terra', 'https://developers.openai.com/api/docs/models/gpt-5.6-terra'],
   ['gpt-5.6-luna', 'https://developers.openai.com/api/docs/models/gpt-5.6-luna'],
diff --git a/src/lib/pricing.mjs b/src/lib/pricing.mjs
index 873df4ff..1311e1bc 100644
--- a/src/lib/pricing.mjs
+++ b/src/lib/pricing.mjs
@@ -6,9 +6,9 @@
 //
 // The cache multipliers are the whole point of this module. On a real corpus
 // ~96% of tokens are cache reads, which bill at 0.1× input for nearly every
-// model (0.025× on Fable 5.1 / Mythos 5.1 — a per-entry override, see below);
-// pricing them as fresh input overstates cost by roughly 10×, or 40× on those
-// two. Cache-write premiums apply to Anthropic writes (1.25× for the 5-minute
+// model (0.025× on Fable 5.1 / Mythos 5.1, 0.05× on Opus 5.5 — per-entry
+// overrides, see below); pricing them as fresh input overstates cost by
+// roughly 10×, or 20–40× on those three. Cache-write premiums apply to Anthropic writes (1.25× for the 5-minute
 // tier, 2× for the 1-hour tier) and OpenAI GPT-5.6 and later; older OpenAI
 // models use the ordinary input rate.
 //
@@ -19,7 +19,7 @@
 // ── Table ────────────────────────────────────────────────────────────────────
 
 /** The baseline verification date; newer entries may carry their own asOf — date-stamped in the UI. */
-export const PRICES_AS_OF = '2026-09-08';
+export const PRICES_AS_OF = '2026-09-23';
 
 // ── Rate constructors ────────────────────────────────────────────────────────
 // A rate entry is always a SCHEDULE — an ordered list of periods, each with the
@@ -87,7 +87,13 @@ export const PRICES = {
   'claude-mythos-5-1': anthropic(10, 50, { cacheReadMultiplier: 0.025, asOf: '2026-09-08' }),
   'claude-fable-5': anthropic(10, 50),
   'claude-mythos-5': anthropic(10, 50),
-  // Anthropic — Opus line (5 and the prior generations share a price)
+  // Anthropic — Opus line. Opus 5.5 (released 2026-09-22) is CHEAPER than the
+  // generations before it, $4/$20, and reads cache at 0.05x base input ($0.20),
+  // not 0.1x (platform.claude.com/docs/en/about-claude/pricing, verified
+  // 2026-09-23). It needs its own key: without one, `claude-opus-5-5` matches
+  // `claude-opus-5` on the token boundary and is priced as Opus 5.
+  'claude-opus-5-5': anthropic(4, 20, { cacheReadMultiplier: 0.05, asOf: '2026-09-23' }),
+  // Opus 5 and the prior generations share a price.
   'claude-opus-5': anthropic(5, 25),
   'claude-opus-4-8': anthropic(5, 25),
   'claude-opus-4-7': anthropic(5, 25),
@@ -102,7 +108,7 @@ export const PRICES = {
   'claude-haiku-4-5': anthropic(1, 5),
 
   // OpenAI (Codex). Verified against the individual OpenAI API model pages on
-  // 2026-09-08. There is NO machine-readable pricing in
+  // 2026-09-23. There is NO machine-readable pricing in
   // ~/.codex/models_cache.json (checked: zero price/pricing/usd keys), so this
   // table is maintained by hand and is the most drift-prone thing in this file.
   //
@@ -125,6 +131,12 @@ export const PRICES = {
   // https://developers.openai.com/api/docs/models/gpt-6-astra (2026-09-08).
   // Standard rates: input $10, cached input $1, writes $12.50, output $50 / MTok.
   'gpt-6-astra': openai(10, 50, { cacheWriteMultiplier: 1.25 }),
+  // GPT-6 Sol and Luna (released 2026-09-22; the model pages' "release date"
+  // field shows the knowledge cutoff instead). Standard rates, cached input 0.1x,
+  // writes 1.25x, same >272K surcharge as Astra (unmodelled):
+  // https://developers.openai.com/api/docs/models/gpt-6-sol and .../gpt-6-luna.
+  'gpt-6-sol': openai(2, 10, { cacheWriteMultiplier: 1.25, asOf: '2026-09-23' }),
+  'gpt-6-luna': openai(0.1, 0.5, { cacheWriteMultiplier: 1.25, asOf: '2026-09-23' }),
   'gpt-5.6-sol': openai(4, 20, { cacheWriteMultiplier: 1.25 }),
   'gpt-5.6-terra': openai(2, 12, { cacheWriteMultiplier: 1.25 }),
   'gpt-5.6-luna': openai(0.2, 1.2, { cacheWriteMultiplier: 1.25 }),
@@ -191,10 +203,14 @@ export const CACHE_WRITE_1H_MULTIPLIER = 2;
 
 // Model ids arrive with inconsistent separators and case (`claude-opus-4-8` vs
 // `claude-opus-4.8`, Codex's `gpt-5.6`), so both key and id are normalised the
-// same way before comparison. Underscores and slashes fold too, so a namespaced
-// id (`anthropic/claude-opus-5`) still ends on the same tokens.
+// same way before comparison.
 const normalize = (s) => String(s).toLowerCase().replace(/[._/]/g, '-');
 
+// A provider-namespaced id (`anthropic/claude-opus-4.7`, `openai/gpt-5.5`, as
+// OpenCode and custom Codex providers record them) is matched on its last path
+// segment. Matching is by prefix, so the namespace would otherwise hide the model.
+const bareModel = (s) => String(s).slice(String(s).lastIndexOf('/') + 1);
+
 // A prefix only counts on a TOKEN boundary: `claude-opus-5` must match
 // `claude-opus-5-20260401` but not `claude-opus-50`, which is a different model.
 const isPrefixOf = (key, id) => id === key || (id.startsWith(key) && id[key.length] === '-');
@@ -239,7 +255,7 @@ function periodOn(periods, day) {
  * never throws.
  *
  * `cacheReadMultiplier` defaults to the module-wide `CACHE_READ_MULTIPLIER`
- * unless the matched entry carries its own (Claude 5.1 or OpenAI Pro) —
+ * unless the matched entry carries its own (Claude 5.1, Opus 5.5 or OpenAI Pro) —
  * see the `schedule()` comment in the table above.
  *
  * `day` (ISO `YYYY-MM-DD`) selects the rate IN EFFECT ON THAT DAY. Cost
@@ -250,7 +266,7 @@ function periodOn(periods, day) {
  * Omitting `day` prices as of `PRICES_AS_OF` (see `periodOn`).
  */
 export function priceFor(model, provider, day) {
-  const normalized = typeof model === 'string' ? normalize(model) : '';
+  const normalized = typeof model === 'string' ? normalize(bareModel(model)) : '';
   // Official exact alias; unknown gpt-5.6-* variants must still remain unknown.
   const id = normalized === 'gpt-5-6' ? 'gpt-5-6-sol' : normalized;
   if (id) {
@@ -289,7 +305,8 @@ const tokens = (v) => (Number.isFinite(v) && v > 0 ? v : 0);
  * every write at the 5-minute rate, exactly as before the split was retained.
  *
  * `cacheReadMultiplier` is 0.1 for nearly every model but is resolved per-model
- * via `priceFor` (Fable 5.1 / Mythos 5.1 price cache reads at 0.025x — see the
+ * via `priceFor` (Fable 5.1 / Mythos 5.1 price cache reads at 0.025x, Opus 5.5
+ * at 0.05x — see the
  * `PRICES` table comment), so this is never hardcoded here.
  *
  * All counters are optional and default to 0, so all-zero usage returns exactly
diff --git a/src/lib/routing.mjs b/src/lib/routing.mjs
index 56fc7af8..3b3a8b9a 100644
--- a/src/lib/routing.mjs
+++ b/src/lib/routing.mjs
@@ -77,7 +77,7 @@ export const SUBSCRIPTION_PROVIDERS = new Set(['claude-code', 'codex', 'ollama',
 // WITHOUT saying "per-token" gets read as cost-per-task, which is the axis users
 // actually pay on. No workload benchmark is embedded in this catalog; notes
 // describe curated roles, not measured superiority.
-export const MODEL_CATALOG_VERIFIED = '2026-09-08';
+export const MODEL_CATALOG_VERIFIED = '2026-09-23';
 export const COST_AXIS_NOTE = 'per-token price ≠ per-task cost — compare total tokens, cache use and agentic turns on representative tasks';
 // Tier names are the pairing key for swapHostModel(): a codex tier only mirrors
 // to a claude model (and back) when BOTH catalogs use the same tier string.
@@ -85,7 +85,10 @@ export const COST_AXIS_NOTE = 'per-token price ≠ per-task cost — compare tot
 // renaming one side silently degrades every mirrored route to cat[0].
 export const MODEL_CATALOG = {
   claude: [
-    { id: 'claude-opus-5', tier: 'reasoning', note: 'reasoning-oriented preset; compare agentic turns and per-task cost on representative work' },
+    // Opus 5.5 (released 2026-09-22) is the reasoning preset: Anthropic's
+    // recommended default for most workloads, at a lower per-token rate than
+    // Opus 5 and 0.05x cache reads. Opus 5 stays listed for user pins.
+    { id: 'claude-opus-5-5', tier: 'reasoning', note: 'reasoning-oriented preset; compare agentic turns and per-task cost on representative work' },
     { id: 'claude-sonnet-5', tier: 'balanced', note: 'balanced preset for review, specification and release work' },
     { id: 'claude-fable-5-1', tier: 'flagship', note: 'flagship preset for demanding tasks; availability depends on the host and account' },
     { id: 'claude-haiku-4-5-20251001', tier: 'fast', note: 'fast-tier preset for high-volume mechanical work' },
@@ -93,14 +96,25 @@ export const MODEL_CATALOG = {
     // longer what ak routes to by default. Kept listed so divergedRoutes can name
     // its cost-per-task trade when a policy is still pointing at it.
     { id: 'claude-opus-4-8', tier: 'prior', note: 'prior-generation option retained for user pins; compare measured per-task results before switching' },
+    { id: 'claude-opus-5', tier: 'prior', note: 'prior reasoning preset retained for user pins; compare measured per-task results before switching' },
     { id: 'claude-mythos-5-1', tier: 'restricted', note: 'restricted-access option; entitlement must be established independently' },
     { id: 'claude-fable-5', tier: 'prior', note: 'prior flagship option retained for explicit selection' },
   ],
   codex: [
-    { id: 'gpt-5.6-sol', tier: 'flagship', note: 'flagship preset for complex coding workflows' },
-    { id: 'gpt-5.6-terra', tier: 'balanced', note: 'balanced preset for everyday implementation and testing' },
-    { id: 'gpt-5.6-luna', tier: 'fast', note: 'fast-tier preset for mechanical work, documentation and packaging' },
-    { id: 'gpt-6-astra', tier: 'frontier', note: 'frontier preset; inspect current model evidence and prices before selection' },
+    // GPT-6 class (Sol/Luna released 2026-09-22; Astra 2026-09-03). Codex's own
+    // catalog ranks these first and labels every GPT-5.6 model "Older", and
+    // OpenAI's Codex docs make Sol the default preset
+    // (learn.chatgpt.com/docs/models, verified 2026-09-23). Tier pairing with the
+    // claude catalog follows role and per-token price: Sol ↔ Sonnet 5 ($2/$10
+    // both), Astra ↔ Fable 5.1 ($10/$50 both), Luna ↔ Haiku. Sol is first, so a
+    // claude tier with no codex twin (reasoning) mirrors to the workhorse.
+    { id: 'gpt-6-sol', tier: 'balanced', note: 'workhorse preset for coding, testing and everyday work' },
+    { id: 'gpt-6-luna', tier: 'fast', note: 'fast-tier preset for mechanical work, documentation and packaging' },
+    { id: 'gpt-6-astra', tier: 'flagship', note: 'frontier preset for the hardest end-to-end work; compare per-task cost before selection' },
+    // Still served, retained for user pins. Codex labels them "Older".
+    { id: 'gpt-5.6-sol', tier: 'prior', note: 'prior coding preset retained for user pins; compare measured per-task results before switching' },
+    { id: 'gpt-5.6-terra', tier: 'prior', note: 'prior balanced preset retained for user pins' },
+    { id: 'gpt-5.6-luna', tier: 'prior', note: 'prior fast preset retained for user pins' },
   ],
 };
 
@@ -239,16 +253,16 @@ export function swapRoute(route) {
 const R = (host, model, escalation) => ({ host, model, ...(escalation ? { escalation } : {}) });
 export const DEFAULT_ROUTES = {
   specification:       R('claude', 'claude-sonnet-5'),
-  architecture:        R('claude', 'claude-opus-5'),
-  design:              R('claude', 'claude-opus-5'),
-  implementation:      R('codex',  'gpt-5.6-terra', [{ host: 'claude', model: 'claude-opus-5' }]),
-  testing:             R('codex',  'gpt-5.6-terra', [{ host: 'claude', model: 'claude-opus-5' }]),
+  architecture:        R('claude', 'claude-opus-5-5'),
+  design:              R('claude', 'claude-opus-5-5'),
+  implementation:      R('codex',  'gpt-6-sol', [{ host: 'claude', model: 'claude-opus-5-5' }]),
+  testing:             R('codex',  'gpt-6-sol', [{ host: 'claude', model: 'claude-opus-5-5' }]),
   review:              R('claude', 'claude-sonnet-5'),
-  'security-scan':     R('codex',  'gpt-5.6-terra'),
-  'security-analysis': R('claude', 'claude-opus-5'),
-  documentation:       R('codex',  'gpt-5.6-luna'),
-  debugging:           R('claude', 'claude-opus-5'),
-  packaging:           R('codex',  'gpt-5.6-luna'),
+  'security-scan':     R('codex',  'gpt-6-sol'),
+  'security-analysis': R('claude', 'claude-opus-5-5'),
+  documentation:       R('codex',  'gpt-6-luna'),
+  debugging:           R('claude', 'claude-opus-5-5'),
+  packaging:           R('codex',  'gpt-6-luna'),
   release:             R('claude', 'claude-sonnet-5'),
 };
 
diff --git a/tests/kit/dashboard-integration-identity.test.mjs b/tests/kit/dashboard-integration-identity.test.mjs
index b414c8c1..c033ca9a 100644
--- a/tests/kit/dashboard-integration-identity.test.mjs
+++ b/tests/kit/dashboard-integration-identity.test.mjs
@@ -1,6 +1,7 @@
 import { test } from 'node:test';
 import assert from 'node:assert/strict';
 import { routingPayload } from '../../src/lib/dashboard-server.mjs';
+import { DEFAULT_ROUTES } from '../../src/lib/routing.mjs';
 
 test('dashboard routing describes host assignment without manufacturing a provider', () => {
   const payload = routingPayload({
@@ -58,7 +59,7 @@ test('retirement and divergence are separate signals on the wire', () => {
   const route = payload.routes.find((entry) => entry.activity === 'architecture');
   assert.equal(route.model, 'claude-opus-4-8', 'a superseded-but-live model is left alone');
   assert.equal(Object.hasOwn(route, 'retiredFrom'), false, 'not a retirement');
-  assert.equal(route.diverged.defaultModel, 'claude-opus-5');
+  assert.equal(route.diverged.defaultModel, DEFAULT_ROUTES.architecture.model);
   assert.ok(route.diverged.currentNote && route.diverged.defaultNote,
     'both sides must carry their cost-per-task note so the trade is legible');
 });
diff --git a/tests/kit/model-dashboard-read-model.test.mjs b/tests/kit/model-dashboard-read-model.test.mjs
index f991c540..042c891e 100644
--- a/tests/kit/model-dashboard-read-model.test.mjs
+++ b/tests/kit/model-dashboard-read-model.test.mjs
@@ -656,7 +656,7 @@ test('Astra inventory and unlinked routes show first-party pricing and its verif
   const projected = createDashboardModelPayload(input, { key: KEY }).snapshot.models[0];
   assert.equal(projected.pricing.input, 10);
   assert.equal(projected.pricing.output, 50);
-  assert.equal(projected.pricing.asOf, '2026-09-08');
+  assert.equal(projected.pricing.asOf, '2026-09-23');
   assert.equal(projected.pricing.sourceUrl, 'https://developers.openai.com/api/docs/models/gpt-6-astra');
   assert.equal(projected.dimensions.entitled.value, null);
   input.snapshot.models = [];
diff --git a/tests/kit/model-discovery-claude-codex.test.mjs b/tests/kit/model-discovery-claude-codex.test.mjs
index 2aae6c8b..433053b9 100644
--- a/tests/kit/model-discovery-claude-codex.test.mjs
+++ b/tests/kit/model-discovery-claude-codex.test.mjs
@@ -54,11 +54,11 @@ test('Claude keeps the 1M selector as a variant instead of a duplicate base mode
 
 test('Claude public facts retain first-party lifecycle, discovery, limits, and scope', () => {
   const result = discoverAnthropicPublicCatalog({
-    capturedAt: '2026-09-02T13:00:00.000Z', scope: { profile: 'default' }, scopeKey: SCOPE_KEY,
+    capturedAt: '2026-09-23T13:00:00.000Z', scope: { profile: 'default' }, scopeKey: SCOPE_KEY,
   });
   assert.equal(result.source.id, 'anthropic-docs');
   assert.equal(result.source.ownerType, 'provider');
-  assert.equal(result.source.sourceVersion, '2026-09-02');
+  assert.equal(result.source.sourceVersion, '2026-09-23');
   assert.equal(result.models.length, ANTHROPIC_PUBLIC_MODELS.length);
 
   const fable = result.models.find((model) => model.identity.modelId === 'claude-fable-5');
@@ -123,7 +123,7 @@ test('Claude public facts include Fable 5.1 and Mythos 5.1 as active, priced ent
 
 test('Claude bundled public facts become explicitly stale instead of silently aging', () => {
   const result = discoverAnthropicPublicCatalog({
-    capturedAt: '2026-12-09T00:00:00.000Z', scopeKey: SCOPE_KEY,
+    capturedAt: '2026-12-23T00:00:01.000Z', scopeKey: SCOPE_KEY, // just past 90 days after 2026-09-23
   });
   assert.equal(result.source.status, 'stale');
   assert.equal(result.models[0].evidence.every(({ freshness }) => freshness === 'stale'), true);
@@ -207,3 +207,10 @@ test('Codex inventory derives an effective override from a valid per-model clamp
     configRaw: 'model_context_window = 872000\n', scopeKey: SCOPE_KEY });
   assert.equal(result.models[0].variant.effectiveContextWindow, 828400);
 });
+
+test('Claude public facts include Opus 5.5 as an active 1M-context entry at its own rate', () => {
+  const opus55 = ANTHROPIC_PUBLIC_MODELS.find((model) => model.id === 'claude-opus-5-5');
+  assert.ok(opus55, 'Opus 5.5 is catalogued');
+  assert.deepEqual([opus55.lifecycle, opus55.contextLimit, opus55.outputLimit, opus55.retirementNotBefore],
+    ['active', 1_000_000, 128_000, '2027-09-22']);
+});
diff --git a/tests/kit/pricing.test.mjs b/tests/kit/pricing.test.mjs
index 6b2dc660..e35496c8 100644
--- a/tests/kit/pricing.test.mjs
+++ b/tests/kit/pricing.test.mjs
@@ -10,7 +10,7 @@ const M = 1_000_000;
 // ── Table shape ──────────────────────────────────────────────────────────────
 
 test('PRICES_AS_OF is the ISO date the table was last verified', () => {
-  assert.equal(PRICES_AS_OF, '2026-09-08');
+  assert.equal(PRICES_AS_OF, '2026-09-23');
 });
 
 test('every PRICES entry carries finite in/out rates, a provider, and asOf', () => {
@@ -220,7 +220,7 @@ test('Astra resolves its published rates and independent verification date', ()
     const p = priceFor(id, 'openai');
     assert.equal(p.matched, true);
     assert.equal(p.key, 'gpt-6-astra');
-    assert.deepEqual([p.in, p.out, p.cacheReadMultiplier, p.asOf], [10, 50, 0.1, '2026-09-08']);
+    assert.deepEqual([p.in, p.out, p.cacheReadMultiplier, p.asOf], [10, 50, 0.1, PRICES_AS_OF]);
   }
   assert.equal(priceFor('gpt-6-astral').matched, false);
   assert.equal(priceFor('gpt-6').matched, false);
@@ -259,3 +259,42 @@ test('OpenAI cache premiums follow the published model generation', () => {
   assert.equal(costOf({ model: 'gpt-5.6', cacheWrite: M }), 5);
   assert.equal(costOf({ model: 'gpt-6-astra', cacheWrite: M }), 12.5);
 });
+
+
+// Opus 5.5 (2026-09-22) is priced BELOW the Opus line before it and reads cache
+// at 0.05x. Before its own key existed, `claude-opus-5-5` matched `claude-opus-5`
+// on the token boundary and was costed at $5/$25 with 0.1x cache reads.
+test('Opus 5.5 resolves its own row: $4/$20, 0.05x cache reads, 1.25x/2x writes', () => {
+  const p = priceFor('claude-opus-5-5');
+  assert.equal(p.key, 'claude-opus-5-5', 'must not fall through to claude-opus-5');
+  assert.deepEqual([p.in, p.out, p.cacheReadMultiplier, p.asOf], [4, 20, 0.05, '2026-09-23']);
+  for (const [counter, expected] of Object.entries({ input: 4, output: 20, cacheRead: 0.2, cacheWrite: 5 })) {
+    assert.equal(costOf({ model: 'claude-opus-5-5', [counter]: M }), expected, counter);
+  }
+  assert.equal(costOf({ model: 'claude-opus-5-5', cacheWrite: M, cacheWrite1h: M }), 8, '1-hour writes at 2x');
+  assert.equal(priceFor('claude-opus-5').key, 'claude-opus-5', 'Opus 5 keeps its own rate');
+  assert.equal(priceFor('claude-opus-5-20260401').key, 'claude-opus-5', 'a dated Opus 5 id is not Opus 5.5');
+});
+
+test('GPT-6 Sol and Luna carry their published standard rates and write premium', () => {
+  for (const [model, rates] of Object.entries({
+    'gpt-6-sol': { input: 2, output: 10, cacheRead: 0.2, cacheWrite: 2.5 },
+    'gpt-6-luna': { input: 0.1, output: 0.5, cacheRead: 0.01, cacheWrite: 0.125 },
+  })) {
+    assert.equal(priceFor(model).key, model);
+    for (const [counter, expected] of Object.entries(rates)) {
+      assert.ok(Math.abs(costOf({ model, [counter]: M }) - expected) < 1e-12, `${model} ${counter}`);
+    }
+  }
+  assert.equal(priceFor('gpt-5.6-sol').key, 'gpt-5.6-sol', 'the GPT-5.6 namesake keeps its own rate');
+});
+
+// OpenCode and custom Codex providers record provider-namespaced ids. Matching
+// is by prefix, so before this the namespace sent them to the fallback rate.
+test('a provider-namespaced id is matched on its model segment', () => {
+  assert.equal(priceFor('anthropic/claude-opus-4.7').key, 'claude-opus-4-7');
+  assert.equal(priceFor('openai/gpt-5.5').key, 'gpt-5.5');
+  assert.equal(priceFor('openrouter/anthropic/claude-sonnet-5').key, 'claude-sonnet-5');
+  assert.equal(priceFor('anthropic/unknown-model').matched, false);
+  assert.equal(priceFor('anthropic/').matched, false);
+});
diff --git a/tests/kit/routing-divergence.test.mjs b/tests/kit/routing-divergence.test.mjs
index 04948a04..31c1d3bd 100644
--- a/tests/kit/routing-divergence.test.mjs
+++ b/tests/kit/routing-divergence.test.mjs
@@ -66,7 +66,7 @@ test('divergedRoutes returns [] for an empty policy (nothing persisted, nothing
 
 test('divergedRoutes reports a seeded entry whose model no longer matches the default', () => {
   // The exact #55 reproduction: a machine seeded pre-alpha.22 still pinned to
-  // the prior Opus generation while DEFAULT_ROUTES moved to claude-opus-5.
+  // the prior Opus generation while DEFAULT_ROUTES moved on.
   const policy = { architecture: { host: 'claude', model: 'claude-opus-4-8', provenance: 'seeded' } };
   const out = divergedRoutes(policy);
   assert.equal(out.length, 1);
@@ -121,12 +121,13 @@ test('the codex-primary clean-seed assertion actually covers a LADDER-bearing ac
 test('a corrupted rung on a MIRRORED activity is still reported (mirroring is not amnesty)', () => {
   const seed = seedActivityRoutes({ primary: 'codex' });
   const [act] = Object.entries(seed).find(([, r]) => r.escalation?.length);
+  const mirroredRung = seed[act].escalation[0].model;
   seed[act].escalation[0].model = 'gpt-5.6-luna';
   const [d] = divergedRoutes(seed);
   assert.equal(d.activity, act);
   assert.equal(d.modelDiverged, false, 'only the rung moved — the primary model is untouched');
   assert.equal(d.escalation[0].model, 'gpt-5.6-luna');
-  assert.equal(d.escalation[0].defaultModel, 'gpt-5.6-sol', 'compared against the MIRRORED default rung');
+  assert.equal(d.escalation[0].defaultModel, mirroredRung, 'compared against the MIRRORED default rung');
 });
 
 test('divergedRoutes ignores a seeded entry that still matches the default', () => {
@@ -138,21 +139,26 @@ test('divergedRoutes ignores a seeded entry that still matches the default', ()
 });
 
 /** The pre-alpha.22 seed: a REAL seeded policy (escalation ladders included,
- *  exactly as seedActivityRoutes writes them) with every claude-opus-5 pin rewound
+ *  exactly as seedActivityRoutes writes them) with every default-Opus pin rewound
  *  to claude-opus-4-8 — primary models and escalation rungs alike. Building this
  *  from seedActivityRoutes rather than by hand matters: a fixture that silently
  *  dropped `escalation` would under-report divergence and hide the escalation-only
  *  case entirely. */
 function priorCatalogSeed({ primary = 'claude' } = {}) {
   // Two rewind rules, one per primary. A claude-primary seed carries
-  // claude-opus-5; a codex-primary seed mirrors every claude model away, so it
+  // the default Opus; a codex-primary seed mirrors every claude model away, so it
   // carries none — rewinding only that id would make the mirrored fixture
   // silently vacuous (it tested nothing at all once the catalog's tier pairing
   // became complete). gpt-5.6-sol appears ONLY in the mirrored seed, so the
   // second rule bites exactly where the first cannot. Both targets are current,
   // non-retired models: this fixture is about divergence, and letting a RETIRED
   // id in would conflate it with the substitution mechanism.
-  const PRIOR = { 'claude-opus-5': 'claude-opus-4-8', 'gpt-5.6-sol': 'gpt-5.6-terra' };
+  // Keyed on the CURRENT defaults so the fixture follows a future default change
+  // instead of silently rewinding nothing. The claude default appears only in a
+  // claude-primary seed, the codex workhorse is the rung a mirrored seed carries.
+  const PRIOR = primary === 'codex'
+    ? { [DEFAULT_ROUTES.implementation.model]: 'gpt-5.6-sol' }
+    : { [DEFAULT_ROUTES.architecture.model]: 'claude-opus-4-8' };
   const rewind = (m) => PRIOR[m] ?? m;
   const policy = seedActivityRoutes({ hosts: ['claude', 'codex'], primary });
   const out = {};
diff --git a/tests/kit/status-viability.test.mjs b/tests/kit/status-viability.test.mjs
index db528ce0..c40cd83f 100644
--- a/tests/kit/status-viability.test.mjs
+++ b/tests/kit/status-viability.test.mjs
@@ -91,7 +91,7 @@ const cfgWithChain = (aqeFallback) => ({
  *  writes them: two of the six diverge ONLY on their escalation rung, and a
  *  hand-built fixture that dropped `escalation` would silently under-report. */
 function divergedPolicy() {
-  const rewind = (m) => (m === 'claude-opus-5' ? 'claude-opus-4-8' : m);
+  const rewind = (m) => (m === DEFAULT_ROUTES.architecture.model ? 'claude-opus-4-8' : m);
   const seed = seedActivityRoutes({ hosts: ['claude', 'codex'] });
   const activityRoutes = {};
   for (const [act, r] of Object.entries(seed)) {