For most thunks, choose a stable model. These are our recommended models for everyday use. You can also try experimental models for early access to newer options.
We test every model before making it available. Learn more about our testing.
Model status
Status | What it means for you | What to do |
Stable | Our recommended models for everyday use. | Choose one for most thunks. |
Experimental | Available to try while we continue evaluating it. | Test it on a few examples before relying on it for important work. |
Deprecated | Being phased out, but still available. | Switch to a stable or experimental model and check the results before retirement. |
Retired | No longer available. | Check which replacement your thunk is using, or choose another model if needed. |
In the model picker, experimental models are labeled (experimental) and deprecated models are labeled (deprecating soon). Stable models have no label.
Supported models
Last updated: September 23, 2026.¹
Stable
If you're not sure which to choose, start with gpt-5.1, the default model.
OpenAI: gpt-5.1 (default), gpt-5-mini, gpt-5.4, gpt-5.4-mini, gpt-5.6-luna, gpt-5.6-terra
Google: gemini-3.5-flash, gemini-3.7-flash
Experimental
Google: gemini-3.8-flash
Other providers: command-a-plus, kimi-k2-thinking, kimi-k3, mistral-medium-3, muse-spark-1.3
Deprecated
Model | Retirement date |
gpt-4.1 | Not yet announced |
gpt-4.1-mini | Not yet announced |
gemini-3.0-flash | Not yet announced |
gemini-2.5-pro | Not yet announced |
glm-5.2 | Not yet announced |
When a model is phased out
We update this article when a model is being phased out or scheduled for retirement. Check here for the latest status and any announced retirement dates.
Deprecated models
You can keep using a deprecated model until it is retired. We announce retirement dates in this article. Before that date, choose a replacement and check a few familiar examples. For most thunks, we recommend a stable model.
Retired models
When a model is retired, your thunk automatically uses a supported replacement from the same model family, if one is available. A message in the thunk's chat tells you about the change. If no replacement is available, you'll need to choose another model before the thunk can run.
Results may change when you switch models. We recommend switching before retirement and checking a few familiar examples to make sure the results meet your needs. For a step-by-step way to do that with your test work items, see Move a thunk to a new AI model.
¹ These lists apply to app.thunk.ai. If your organization runs its own version of Thunk.AI, the available models may differ. We will update this page when models are retired.