Model table
The model identities and public surfaces covered by the current research boundary.
Monitored model identities
This is a coverage table, not a leaderboard. It records which public surfaces must remain distinct when interpreting reports.
| Model | Provider | Observed surfaces | Interpretation boundary |
|---|---|---|---|
| GPT-5.6 Sol | OpenAI | Hosted API, Codex and product surfaces | Public records distinguish outages, routing and product changes from capability evidence. |
| GPT-5.6 Terra | OpenAI | Hosted API tier | Price and serving changes are not treated as weight-level degradation. |
| GPT-5.6 Luna | OpenAI | Hosted API tier | Public comparisons are snapshots rather than a controlled time series. |
| Claude Fable 5 | Anthropic | Hosted model with documented safeguards and fallback behavior | Access and fallback configuration complicate longitudinal comparison. |
| Claude Mythos 5 | Anthropic | Restricted access tier | Availability and access policy changes are tracked separately. |
| Claude Opus 5 | Anthropic | Hosted API and product surfaces | Independent comparison evidence is dated and configuration-sensitive. |
| Grok 4.5 | xAI | API and Grok product surfaces | Product changelog and status evidence do not expose one universal snapshot. |
| GLM-5.2 | Z.ai | Open weights and hosted surfaces | Public model artifacts are separated from hosted evaluation configuration. |
| DeepSeek V4 Pro | DeepSeek | Open weights and API | Legacy alias routing is not attributed to Pro capability decline. |
| Kimi K3 | Moonshot AI | Open weights and hosted products | Release and independent snapshots are preserved with their dates. |