Model coverage and evidence (v4.0)
Reviewed: September 27, 2026. Previous version: v3.0, July 6, 2026 (commit bffa2a5). This reference tracks general-purpose text models and relevant product updates after that baseline. It is a dated snapshot, not a promise of future model coverage or identical access on every plan.
Read this file when a request names a model, asks which versions are covered, or needs release evidence. Ordinary writing can use the core skill without loading this timeline.
Verified release timeline
ChatGPT / OpenAI
| Date (2026) | Model or update | Scope and source |
|---|---|---|
| July 9 | GPT-5.6 Sol, Terra, Luna | General availability across ChatGPT, Codex and API; access differs by tier and surface. GPT-5.6 announcement |
| August 6 | GPT-5.6 Sol ChatGPT update; Luna access expansion | A product update within the generation, not a new base model. ChatGPT update |
| September 3 | GPT-6 Astra | API release and an initially limited organizational ChatGPT rollout. API changelog, ChatGPT release notes |
| September 22 | GPT-6 Sol, Luna | Text-output API models, also in ChatGPT Work and Codex. Their availability there does not establish a default in ordinary ChatGPT Chat. API changelog, ChatGPT release notes |
GPT-6 Pro is a ChatGPT offering powered by Astra, not another base model. Check the current plan and surface rather than assuming a model from the ChatGPT brand. ChatGPT model/access guide
Writing evidence: OpenAI describes the August Sol update as more focused and says extra reasoning effort should not change the writing tone. That is vendor-described behavior, not an independent prose fingerprint. Update announcement
Claude / Anthropic
| Date (2026) | Model | Scope and source |
|---|---|---|
| July 24 | Claude Opus 5 | Released for Claude and developer use. Opus 5 announcement |
| September 1 | Claude Fable 5.1, Mythos 5.1 | Fable is generally available; Mythos has restricted trusted access. They share an underlying model with different safeguards. The newsroom supplies the exact date; the launch page displays only the month. Launch, Newsroom |
| September 22 | Claude Opus 5.5 | Released. Sonnet 5.5 and Haiku 5.5 were announced as upcoming, not released by this review date. Opus 5.5 announcement |
Writing evidence: Anthropic reports clearer, more natural communication and better adherence to writing instructions for Opus 5.5. This does not validate the old v3 punctuation ratios or a universal “Philosopher” dialect for current Claude models. Opus 5.5 communication examples
Gemini / Google
| Date (2026) | Model | Scope and source |
|---|---|---|
| July 21 | Gemini 3.6 Flash, 3.5 Flash-Lite | General text-capable models. Launch |
| August 13 | Gemini 3.7 Flash | Next Flash release. Launch |
| September 2 | Gemini 3.8 Flash | Latest general-purpose stable Flash release in this review. Launch |
The API registry confirms these dates and lists stable models separately from previews. Gemini 3.1 Pro remains a preview entry in the current catalog; its name is not evidence of a new post-v3 release. Live, TTS, image, embedding and specialist Cyber releases do not establish ordinary prose behavior. App labels and API IDs can differ. Release/deprecation registry, Model catalog
Grok / xAI
| Date (2026) | Model | Scope and source |
|---|---|---|
| July 16 | Grok 4.5 | Launch for developer/agent surfaces; consumer rollout followed separately. Launch |
| August 12 | Grok 4.6 | Initial launch in Grok Build and Cursor; API availability is documented separately. Launch, API release notes |
| September 21 | Grok 4.7 | Latest general-purpose release reviewed; launch includes Grok Build, Cursor and API. Launch |
These releases describe capabilities and distribution, not a measured sarcastic or “edgy” prose signature. Do not equate the Grok brand with a requested voice or assume every consumer surface uses the newest API model. xAI news and rollout records
Editorial checks for all four families
These are editing heuristics proposed by this skill. They are not findings from a controlled comparison of the listed models. Apply a check only when the draft exhibits the problem:
- Remove assistant preambles, unsolicited follow-up offers and process narration from an article, memo or caption. Keep an explanation of the reasoning when the requested format needs it.
- Cut stock reassurance, praise, motivational hooks and forced jokes. Match the requested voice; sarcasm and empathy can be appropriate when the context earns them.
- Replace abstract claims with supported details. Never invent names, numbers, quotes, citations or lived experiences to make text seem specific.
- Check whether each citation supports its nearby claim, whether information is current, and whether uncertainty survived the rewrite. Retrieved snippets and polished summaries still need verification.
- Merge repeated claims and unnecessary headings or bullets. Keep requested lists, comparisons and templates intact.
- Read for repeated sentence shapes and canned endings. Revise for meaning and flow, not a fixed sentence-length quota, arbitrary list size or punctuation detector score.
The skill edits language and structure. It does not identify authorship, guarantee detector results, or prove that a provider's internal writing process works a particular way. No live comparison of these models was run for this release. For future model updates, verify a dated official release, record its access status, and test actual drafts before adding a claimed model-specific pattern.