Models and releases
Models and versions, what they take in and put out, what they cost, where they run, under what licence, and when they are retired.
On the register
- Adaptive thinking replacing manual budgets watching
- Audio billed per minute beside tokens watching
- Cache-read prices diverging by model watching
- Coding tools retiring models on their own clock watching
- Dateless pinned model ids watching
- Embedding models retired for successors watching
- Fine-tuning job creation restricted watching
- Frontier models behind gated programmes watching
- Image models retired for successors watching
- Long-context price tiers watching
- Maximum output limits rising watching
- Meta's lineup moving from downloadable Llama to served Muse watching
- Older models closed to new users watching
- One-million-token windows as the default watching
- Open weights at frontier scale watching
- Permissive licences on the largest open weights watching
- Price changes announced with a date watching
- Priority and fast tiers priced as multipliers watching
- Regional inference at a premium watching
- Retirement notice periods watching
- Robot policies released with weights or an API watching
- Safeguard changes on models already served watching
- Same model, different outlet, different clock watching
- Sampling parameters withdrawn from APIs watching
- Speech models retired for successors watching
- Task-specific variants of a frontier family watching
- Tier and codename proliferation inside families watching
- Tokenisers changing what a token buys watching
- Training content summaries shipped with releases watching
- Video APIs withdrawn watching
- Whether consumer and API data train the vendor's models by default watching
Everything filed here
- 2026-09-22 SpaceXAI released Grok 4.7 with a 500k-token context window · Models and releases · 1 source
- 2026-09-22 Anthropic released Claude Opus 5.5 with a 1M-token context window · Models and releases · 2 sources