Skip to content

refresh: quarterly card refresh 2026-07-28 - #9

Open
russmiles wants to merge 1 commit into
mainfrom
refresh-2026-07-28
Open

refresh: quarterly card refresh 2026-07-28#9
russmiles wants to merge 1 commit into
mainfrom
refresh-2026-07-28

Conversation

@russmiles

Copy link
Copy Markdown
Contributor

Summary

Quarterly refresh of all 13 model cards. Cards were last researched 2026-04-30 (89 days ago). All cards updated to card_version: 0.2.0, last_researched: 2026-07-28, researcher: scheduled-quarterly-refresh-agent (claude-sonnet-4-6).


⚠ Critical flags requiring maintainer attention

1. Google Gemini 2.5 models — confirmed early API outage

All tier-1 provider documentation URLs for gemini-2.5-pro and gemini-2.5-flash returned HTTP 404 at refresh time (July 28, 2026), consistent with the models having gone offline approximately July 9, 2026 — more than three months before the official retirement date of October 16, 2026. API calls to these models may already fail for end users.

Both cards now carry a ⚠ DEPRECATION NOTICE block and recommend immediate migration to gemini-3.5-pro / gemini-3.5-flash respectively.

2. xAI Grok 4 — retirement in 18 days (August 15, 2026)

grok-4 officially retires August 15, 2026, which is 18 days from the date of this refresh. The card carries a deprecation notice. Successor is Grok 4.5.


Per-card outcomes

Provider Model Changes
Anthropic claude-sonnet-4-6 Legacy status added; max output corrected 64k→128k; extended thinking deprecated on this model
Anthropic claude-opus-4-7 Legacy status added; tool-use token overhead corrected to 675/804 (new tokenizer; was 346/313)
Anthropic claude-haiku-4-5 Frontmatter dates only; confirmed still Latest (no Haiku 5 released)
OpenAI gpt-5 ~50% price cut; ChatGPT retired Feb 13 2026; API snapshot shutdown Dec 11 2026
OpenAI gpt-5-mini ~50% price cut; API snapshot shutdown Dec 11 2026; successor gpt-5.4-mini noted
OpenAI o4-mini o4-mini-deep-research shut down July 23 2026; o4-mini-2025-04-16 snapshot shutdown Oct 23 2026
Google gemini-2.5-pro ⚠ DEPRECATION NOTICE; early 404 outage ~July 9; migrate to gemini-3.5-pro immediately
Google gemini-2.5-flash ⚠ DEPRECATION NOTICE; early 404 outage ~July 9; SFT now GA on Gemini API; Flash-Lite variant added; Live API native audio preview; migrate to gemini-3.5-flash immediately
Meta llama-4 Behemoth confirmed shelved (never released); Meta launched closed Muse Spark product
Mistral mistral-large-3 Fine-tuning now GA via managed Mistral API (was unconfirmed in April 2026)
xAI grok-4 ⚠ DEPRECATION NOTICE; retires August 15 2026 (18 days); successor is Grok 4.5
Cohere command-r-plus Command A+ now Cohere flagship (superseding Command A); deprecated on Oracle OCI (Apr 2026) and Microsoft Azure AI Foundry (May 2026)
Alibaba qwen3-coder Qwen 4 Coder released June 2026 (82% SWE-Bench Verified); qwen3-coder-plus pricing updated to $0.650/$3.25 per 1M tokens

Source access notes

Most provider tier-1 URLs returned HTTP 403 at refresh time. Anthropic's platform.claude.com returned 200. Google's Gemini 2.5 URLs returned 404 (indicating actual model offline, not just access-blocked). Operational updates for blocked providers were verified via web search (tier-4) and noted in each card's Section 9 "Refresh note."


Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants