AI Model Version Status Check | Generated by AI
Good question. I looked into each of those models in your MODEL_MAPPING and checked whether they’re currently (as of late 2025) the latest in their respective series. Here’s what I found — and where there may be newer versions.
Analysis of Your Models
| Model Key | Mapped Model | Is It the Latest? | Notes |
|---|---|---|---|
| claude-opus | anthropic/claude-opus-4 |
Not quite | Anthropic released Claude Opus 4.1 on Aug 5, 2025. (Anthropic) So claude-opus-4 is slightly outdated. |
| claude-sonnet | anthropic/claude-4.5-sonnet |
Yes / No | According to the Claude wiki page, Anthropic announced Sonnet 4.5 on Sep 29, 2025, which is more capable than Sonnet 4. (Wikipedia) If by “claude-sonnet” you meant Sonnet 4, then it’s not the latest; but if you used “4.5”, you’re up-to-date. |
| gemini-flash | google/gemini-2.5-flash |
Probably not | Google’s Gemini 2.5 Flash is GA. (blog.google) But there is now a Gemini 3.0 Pro model as of November 2025. (Wikipedia) Depending on the variant, 2.5 may not be “latest”. |
| gemini-pro | google/gemini-3-pro-preview |
Yes-ish | Gemini 3.0 Pro was just rolled out. (Reddit) Given it’s labeled “preview,” it may be the cutting-edge but might not yet be fully stable. |
| kimi-k2 | moonshotai/kimi-k2 |
Likely yes | I found a paper for Kimi K2 (32B MoE). (arXiv) I did not find a stable Kimi-K3 or K2.1 publicly announced, so K2 is probably the current version. |
| deepseek-v3.2 | deepseek/deepseek-v3.2-exp |
Unclear | I didn’t find reliable public evidence that DeepSeek has a newer stable version beyond v3.2 (or at least not one that’s clearly documented). Without a changelog or version registry, it’s hard to confirm. |
| mistral-medium | mistralai/mistral-medium-3.1 |
Yes | Mistral’s own model list shows Mistral Medium 3.1 as the latest medium model. (Mistral AI) |
| qwen-coder | qwen/qwen3-coder |
Unclear | I didn’t see a clearly newer “Qwen 4-coder” or “Qwen 3.1-coder” in public documentation. Without more info, it’s ambiguous whether 3-coder is still the latest. |
| gpt | openai/gpt-5.1 |
Yes | According to sources, GPT-5.1 was released in November 2025. (Wikipedia) |
| grok-code | x-ai/grok-code-fast-1 |
Unclear | I found no public release notes confirming a newer “grok-code” model beyond version 1. Without an official roadmap or model card, this may still be current. |
| grok-fast | x-ai/grok-4.1-fast |
Unclear | Similar to grok-code: no clear public documentation of a “grok-4.2-fast” or equivalent, so 4.1-fast might still be their latest. |
| glm | z-ai/glm-4.6 |
Unclear | I couldn’t find an official “GLM 4.7” or later from Z-AI. Public information on recent GLM versions is limited, so hard to say. |
| minimax | minimax/minimax-m2 |
Unclear | I didn’t find strong public version history for Minimax. It’s possible M2 is the current or a stable version, but I can’t confirm a newer one. |
| kimi-k2-thinking | moonshotai/kimi-k2-thinking |
Likely yes | There’s no public mention of a “Kimi-K2- thinking 2.0” or similar; K2-thinking appears to be a thinking-mode variant of the K2 model. |
Summary & Recommendations
- Not up-to-date: Your mapping for claude-opus probably should point to
claude-opus-4.1instead ofclaude-opus-4. - Gemini: Depending on your use-case, you might consider switching from Gemini 2.5 Flash to a 3.0 model, especially since 3.0 Pro is out.
- Some models are unclear: For Qwen, Grok, DeepSeek, GLM, and Minimax, the public versioning is either limited or not clearly documented, so you may want to revisit their official repos or documentation.
- Kimi K2: Looks stable and current.
If you like, I can check all of the model keys in your mapping (or even more) and produce a fully up-to-date mapping with the latest versions as of now — do you want me to do that?