0.00
0.00–0.00
Benchmarks / Arena (formerly LMArena)
Reported by Arena (formerly LMArena)
Arena (formerly LMArena)
Estimated effect on avoiding calls to tools that do not exist.
- Results dated
- 28 Sep 2026
- Models
- 46
- Unit
- IPS effect estimate
Full results
| # | Model | Arena Agent: tool grounding · 95% range IPS effect estimate, higher is better |
|---|---|---|
| 1 | Grok 4.6xAI | |
| 1 | GPT-5.4OpenAI |
0.00 0.00–0.00 |
| 1 | GPT-6 AstraOpenAI |
0.00 0.00–0.00 |
| 1 | Grok 4.5xAI |
0.00 0.00–0.00 |
| 1 | GPT-5.5OpenAI |
0.00 0.00–0.00 |
| 1 | GPT-6 LunaOpenAI |
0.00 0.00–0.00 |
| 1 | GPT-5.6 SolOpenAI |
0.00 0.00–0.00 |
| 1 | Grok 4.7xAI |
0.00 0.00–0.00 |
| 1 | GPT-5.6 TerraOpenAI |
0.00 0.00–0.00 |
| 1 | GPT-6 SolOpenAI |
0.00 0.00–0.00 |
| 1 | DeepSeek V4 Pro 0813DeepSeek |
0.00 0.00–0.00 |
| 1 | GPT-5.6 LunaOpenAI |
0.00 0.00–0.00 |
| 1 | GLM 5.3 FlashZ.ai |
0.00 0.00–0.00 |
| 1 | GLM 5.3Z.ai |
0.00 0.00–0.00 |
| 1 | Kimi K3Moonshot AI |
0.00 0.00–0.00 |
| 1 | Claude Fable 5Anthropic |
0.00 0.00–0.00 |
| 1 | GPT-5.5OpenAI |
0.00 0.00–0.00 |
| 1 | GLM 5.2Z.ai |
0.00 0.00–0.00 |
| 1 | Muse Spark 1.3Meta |
0.00 0.00–0.00 |
| 1 | Claude Fable 5.1Anthropic |
0.00 0.00–0.00 |
| 1 | Claude Opus 5Anthropic |
0.00 0.00–0.00 |
| 1 | Muse Spark 1.1Meta |
0.00 0.00–0.00 |
| 1 | Muse Spark 1.2Meta |
0.00 0.00–0.00 |
| 1 | Claude Opus 5Anthropic |
0.00 0.00–0.00 |
| 1 | Gemini 3.6 FlashGoogle |
0.00 0.00–0.00 |
| 1 | Claude Opus 5.5Anthropic |
0.00 0.00–0.00 |
| 1 | Hy4 previewTencent |
0.00 -0.00–0.00 |
| 1 | Gemini 3.1 Pro PreviewGoogle |
-0.00 -0.01–0.00 |
| 24 | Gemini 3.8 FlashGoogle |
0.00 0.00–0.00 |
| 24 | Claude Sonnet 5Anthropic |
0.00 0.00–0.00 |
| 25 | DeepSeek V4.1 FlashDeepSeek |
0.00 0.00–0.00 |
| 27 | Claude Opus 4.8Anthropic |
-0.00 -0.00–0.00 |
| 27 | Gemini 3.7 FlashGoogle |
-0.00 -0.00–0.00 |
| 29 | InklingThinkingmachines |
-0.00 -0.00–0.00 |
| 29 | MiniMax M3MiniMax |
-0.00 -0.00–0.00 |
| 30 | DeepSeek V4 Pro 0423DeepSeek |
-0.00 -0.00–-0.00 |
| 30 | Qwen3.7 MaxAlibaba |
-0.00 -0.00–-0.00 |
| 30 | Inkling SmallThinkingmachines |
-0.00 -0.01–-0.00 |
| 31 | Qwen3.8 27BAlibaba |
-0.00 -0.00–-0.00 |
| 38 | MiMo-V2.5-ProXiaomi |
-0.01 -0.01–-0.01 |
| 38 | Qwen3.7 PlusAlibaba |
-0.01 -0.01–-0.01 |
| 39 | Qwen3.8 Flash Next (arena agent)Alibaba |
-0.01 -0.01–-0.01 |
| 39 | Qwen3.8 Max (0902)Alibaba |
-0.01 -0.01–-0.01 |
| 40 | Solar Pro 4Upstage |
-0.02 -0.03–-0.01 |
| 44 | Mistral Medium 3.5Mistral |
-0.02 -0.03–-0.02 |
| 44 | Hy3Tencent |
-0.03 -0.04–-0.02 |
Models share a rank when their ranges overlap. Results as published by Arena (formerly LMArena); we do not re-run them.
What it measures
Estimated effect on avoiding calls to tools that do not exist.
What it does not measure
Not a success rate: the estimated effect of choosing this model as the agent's orchestrator, against an even mix of the models tested, inside Arena's own harness.
Source
Contains data from the Arena Leaderboard Dataset by Arena, licensed under CC BY 4.0. Licence: Creative Commons Attribution 4.0 International.