Telegram

Best agent model

Long-running tasks with tool calls, where being nearly right at step three ruins step nine.

Right now

Claude Opus 5.5 (Max) from Anthropic currently ranks first for multi-step work with tools, ahead of GPT-6 Astra (Max) from OpenAI. The ranking comes from LMArena and was last published 2026-10-16.

Leader
Claude Opus 5.5 (Max)
Made by
Anthropic
Score
18.9
Source
LMArena

Full ranking Human head-to-head preference votes, Bradley-Terry rating.

Rank# Model Made by Score % Votes
1 Claude Opus 5.5 (Max) Anthropic 18.9 10,551
2 GPT-6 Astra (Max) OpenAI 14.1 13,168
3 Claude Fable 5.1 (Max) Anthropic 14.0 16,374
4 GPT-6.1 Sol (Max) OpenAI 13.6 13,444
5 Gemini 4 Argon (High) Google 13.2 30,489
6 Claude Sonnet 5.5 (Max) Anthropic 13.0 7,560
7 GPT-6 Sol (Max) OpenAI 10.3 7,887
8 Claude Opus 5 (Max) Anthropic 8.7 23,770
9 Claude Fable 5 (High) Anthropic 8.2 42,608
10 GPT-5.6 Sol (xHigh) OpenAI 7.1 36,545
11 Muse Wonder (Max) Meta 4.9 7,690
12 Claude Sonnet 5 (High) Anthropic 4.6 33,276
13 MiMo-V2.6-Pro Open Xiaomi 4.0 27,455
14 Muse Spark 1.3 (Max) Meta 3.8 73,912
15 Gemini 3.8 Flash (High) Google 3.7 37,692
16 Grok 4.7 (xHigh) xAI 3.2 15,127
17 GPT-5.5 (xHigh) OpenAI 3.1 54,331
18 GPT-6 Luna (Max) OpenAI 3.0 163,815
19 DeepSeek-V4.1-Flash (Max) Open DeepSeek 2.7 241,716
20 Hy4 preview Open Tencent 2.7 84,828
21 MiMo-V2.6-Flash Open Xiaomi 2.6 19,620
22 GLM-5.2 (Max) Open Zai 2.3 87,460
23 Kimi K3 (Max) Open Moonshot AI 1.9 135,415
24 GLM-5.3 (Max) Open Zai 1.8 86,541
25 Qwen3.8-Max Alibaba 1.8 68,968

Published by LMArena on 2026-10-16 · copied here · no adjustment applied

Other leaderboards