Best AI models for agentic
Agentic tests measure whether a model can call tools correctly, hold a multi-turn conversation with a simulated customer, and keep searching the web until it finds an answer that is hard to locate. Across 8 models with at least one agentic score, Kimi K3.1 Pro has the highest index at 100.0. Among models you can deploy on OpenLLM Buddy, Kimi K3 leads, ranked 3 overall.
Agentic index
Average of each model's scores relative to the best published score on every test in this category (top = 100). Only tests a model reports are counted, and a model needs at least two of them to be ranked.
- OpenLLM Buddy catalog
- Open weights
- Closed API
- 1Kimi K3.1 ProMoonshot AI · 2 of 4 tests100.0
- 2GPT-6 AstraOpenAI · 2 of 4 tests90.5
- 3Kimi K3Moonshot AI · 2 of 4 tests89.5
- 4DeepSeek V4 ProDeepSeek · 2 of 4 tests88.1
- 5Reflection Beam 1.0Reflection AI · 2 of 4 tests78.2
- 6Gemma 4 31BGoogle · 2 of 4 tests68.9
- 7Gemma 4 26B A4BGoogle · 2 of 4 tests56.6
- 8Nemotron 3 UltraNVIDIA · 2 of 4 tests50.8
11 more models report a single agentic test and are listed in the table below without a rank: DeepSeek V4.1 Flash, GLM 5.3 Flash, Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5, GLM 5.3, Mistral Large 4, gpt-oss-120b, Muse Spark Open, Step 5 Preview.
Raw scores by test
Published percentages. A dot means the model has no published score on that test.
| Model | HLE+tools | τ² | BFCL | BrowseComp |
|---|---|---|---|---|
| Kimi K3.1 Pro | 70.0 | · | · | 92.1 |
| GPT-6 Astra | 57.2 | · | · | 91.5 |
| Kimi K3 | 56.0 | · | · | 91.2 |
| DeepSeek V4 Pro | 60.0 | · | · | 83.4 |
| Reflection Beam 1.0 | 51.1 | · | · | 76.8 |
| Gemma 4 31B | 26.5 | 76.9 | · | · |
| Gemma 4 26B A4B | 17.2 | 68.2 | · | · |
| Nemotron 3 Ultra | 37.4 | · | · | 44.4 |
| DeepSeek V4.1 Flash | 63.9 | · | · | · |
| GLM 5.3 Flash | 55.3 | · | · | · |
| Claude Fable 5.1 | 65.0 | · | · | · |
| Claude Opus 5.5 | 67.7 | · | · | · |
| Claude Sonnet 5.5 | 64.5 | · | · | · |
| Claude Haiku 5.5 | 57.4 | · | · | · |
| GLM 5.3 | 62.5 | · | · | · |
| Mistral Large 4 | · | · | · | 84.5 |
| gpt-oss-120b | 19.0 | · | · | · |
| Muse Spark Open | 61.0 | · | · | · |
| Step 5 Preview | · | · | · | 88.7 |
Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works