Telegram

Best model for coding

Judged on real build tasks where a human picks the better result, not on a self-reported pass rate.

Right now

qwen3.8-max-0902 from Alibaba currently ranks first for writing and fixing code, ahead of Claude Opus 5 Max from Anthropic. The ranking comes from LMArena and was last published 2026-09-01.

Leader
qwen3.8-max-0902
Made by
Alibaba
Score
1691
Source
LMArena

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

Rank# Model Made by Arena score Votes
1 qwen3.8-max-0902 Alibaba 1,691 1,389
2 Claude Opus 5 Max Anthropic 1,688 10,334
3 Kimi K3 Max Open Moonshot AI 1,674 4,544
4 qwen3.8-max Alibaba 1,669 3,219
5 Claude Opus 5 High Anthropic 1,661 10,326
6 Hy4 Preview Open Tencent 1,629 1,368
7 Grok 4.6 High xAI 1,629 1,532
8 Claude Fable 5 Anthropic 1,628 9,066
9 Qwen3.8 Flash Next Open Alibaba 1,620 1,876
10 gpt-5.6-sol-xhigh (codex-harness) OpenAI 1,616 10,684
11 Glm 5.3 Max Open Zai 1,608 2,648
12 Glm 5.3 Flash Open Zai 1,604 1,698
13 Qwen3.8 27b Open Alibaba 1,599 3,256
14 Gemini 3.7 Flash High Google 1,587 3,007
15 Glm 5.2 Max Open Zai 1,585 9,569
16 Deepseek v4 Pro High Open DeepSeek 1,584 3,289
17 Deepseek v4 Flash High Open DeepSeek 1,581 3,928
18 Claude Opus 4.8 High Anthropic 1,563 12,603
19 Claude Opus 4.7 Anthropic 1,557 15,430
20 Claude Opus 4.7 High Anthropic 1,556 15,909
21 Grok 4.5 xAI 1,555 7,126
22 Claude Opus 4.6 High Anthropic 1,546 17,908
23 Claude Opus 4.8 Anthropic 1,540 11,543
24 Muse Spark 1.1 Meta 1,540 7,091
25 Gemini 3.6 Flash High Google 1,538 7,726

Published by LMArena on 2026-09-01 · copied here 2026-09-02 · no adjustment applied

Other leaderboards