BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases9

Arena Coding

Text arena rating on coding prompts.

As of 25 Oct 2026, Gemini 4 Argon leads Arena Coding on BenchLeader with 1552, ahead of Claude Fable 5 at 1551, across 368 model configurations with a published result.

Published by
Arena (LMArena)
Category
Coding
Index weight
0.5
Models
368
Data as of
25 Oct 2026
Newest result
22 Oct 2026

CC BY 4.0 — lmarena-ai/leaderboard-dataset on Hugging Face.

What the test looks like

Pairwise votes on prompts classified as coding questions in the text arena: explanations, snippets and debugging in chat form.

How it is scored

Bradley-Terry rating on coding prompts, with style control.

What to keep in mind

Voters judge the chat answer, not whether the code runs, so it measures perceived helpfulness. For executed code, see SWE-bench, LiveCodeBench or Aider.

Top 15
  1. 1Gemini 4 Argon (high)1552
  2. 2Claude Fable 5 (high)1551
  3. 3Claude Opus 4.7 (high)1551
  4. 4Claude Opus 5.5 (high)1548
  5. 5Claude Opus 4.61547
  6. 6Claude Opus 4.6 (high)1547
  7. 7Claude Opus 4.71546
  8. 8Muse Wonder (max)1545
  9. 9GPT-6 Astra (max)1543
  10. 10Muse Spark 1.2 (xhigh)1541
  11. 11Kimi K3 (max)1539
  12. 12MiMo-V2.6-Pro1538
  13. 13Claude Fable 5.1 (max)1538
  14. 14GPT-6.1 Sol (max)1537
  15. 15Muse Spark 1.3 (max)1536
368 of 368
#
1Gemini 4 Argonhigh155270.72026-10-22
2Claude Fable 5high155164.42026-10-22
3Claude Opus 4.7high155161.02026-10-22
4Claude Opus 5.5high154869.42026-10-22
5Claude Opus 4.6154762.52026-10-22
6Claude Opus 4.6high154762.72026-10-22
7Claude Opus 4.7154663.22026-10-22
8Muse Wondermax154567.52026-10-22
9GPT-6 Astramax154370.42026-10-22
10Muse Spark 1.2xhigh154163.22026-10-22
11Kimi K3max153964.82026-10-22
12MiMo-V2.6-Pro153864.02026-10-22
13Claude Fable 5.1max153870.52026-10-22
14GPT-6.1 Solmax153769.92026-10-22
15Muse Spark 1.3max153667.12026-10-22
16Claude Opus 5high153467.92026-10-22
17GPT-5.6 Solxhigh153466.62026-10-22
18Claude Opus 4.8high153360.72026-10-22
19Muse Spark 1.1153362.52026-10-22
20Claude Opus 4.5high153059.22026-10-22
21Claude Opus 4.8152960.12026-10-22
22Muse Spark152964.32026-10-22
23Claude Opus 5max152868.82026-10-22
24Claude Sonnet 4.6152857.12026-10-22
25Gemini 3.8 Flashhigh152864.42026-10-22
26Claude Sonnet 5.5xhigh152867.92026-10-22
27GLM 5.3 Flash152562.72026-10-22
28Qwen3.8 Maxmax152463.32026-10-22
29Qwen3.7max152460.62026-10-22
30DeepSeek V4.1 Flashmax152361.22026-10-22
31Claude Opus 4.5152357.92026-10-22
32Gemini 3.1 Pro152260.52026-10-22
33GLM 5.3max152163.52026-10-22
34GPT-5.4high152158.92026-10-22
35MiMo-V2.5-Pro152154.92026-10-22
36Gemini 3.7 Flashhigh152063.52026-10-22
37Claude Sonnet 5high152060.12026-10-22
38GPT-6 Solmax151966.42026-10-22
39GPT-5.5high151965.32026-10-22
40Claude Sonnet 4.5high151958.12026-10-22
41Gemini 3.6 Flashhigh151862.02026-10-22
42GPT-5.6 Terraxhigh151862.82026-10-22
43Gemini 3 Pro151760.42026-10-22
44Kimi K2.6151655.32026-10-22
45GPT-5.2151558.02026-10-22
46GLM 5.2max151462.32026-10-22
47Dola Seed 2.0 Pro151458.52026-10-22
48Grok 4.5151459.02026-10-22
49GPT-5.5 Instant151358.02026-10-22
50Qwen3.5 Maxmax1513–2026-10-22
51GLM 5.1151355.92026-10-22
52Claude Sonnet 4.5151253.72026-10-22
53Qwen3.6 Maxmax151262.32026-10-22
54ERNIE 5.11512–2026-10-22
55GPT-5.4151158.12026-10-22
56Claude Opus 4.1thinking151156.42026-10-22
57MiMo-V2.6-Flash151159.72026-10-22
58GPT-5.5151160.42026-10-22
59Grok 4.20thinking151158.62026-10-22
60Grok 4.6high151062.92026-10-22

More coding benchmarks · All LLM benchmarks

Cite as: BenchLeader, “Arena Coding leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/lmarena_text_coding.

Arena Coding: questions

What does Arena Coding measure?
Pairwise votes on prompts classified as coding questions in the text arena: explanations, snippets and debugging in chat form. Scores are reported in Elo-style ratings from pairwise comparisons; higher is better.
Which AI model leads Arena Coding?
Gemini 4 Argon leads Arena Coding with 1552 as of 25 Oct 2026, ahead of Claude Fable 5 at 1551.
How many models have Arena Coding results?
368 model configurations have a Arena Coding result on BenchLeader, all taken from Arena (LMArena).
Who runs Arena Coding and how often is it updated?
Arena Coding is published by Arena (LMArena). BenchLeader re-reads the published results once a day and records the date each result was published.
Does Arena Coding count toward the BenchLeader Index?
Yes. Arena Coding contributes to the coding category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.