pacphi · pacphi · Sep 23, 2026 · Sep 23, 2026
diff --git 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
@@ -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
@@ -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