LMArena Code Leaderboard (September 2026)
In September 2026, GPT-6 Astra ranks first with 1,800, followed by claude-fable-5.1-max (1,758) and claude-opus-5-max (1,687). Top 20 of 121 candidates.
Data as of September 2026
| # | Model | Provider | Score | Votes |
|---|---|---|---|---|
| 1 |  GPT-6 Astra | OpenAI | 1,800 | 2.3K |
| 2 |  claude-fable-5.1-max | Anthropic | 1,758 | 3.0K |
| 3 | claude-opus-5-max | Anthropic | 1,687 | 12.1K |
| 4 |  qwen3.8-max-0902 | Alibaba | 1,681 | 2.3K |
| 5 |  Kimi K3 | Moonshot | 1,679 | 1.8K |
| 6 | kimi-k3-max | Moonshot | 1,674 | 4.5K |
| 7 | Qwen 3.8 Max | Alibaba | 1,671 | 3.2K |
| 8 | claude-opus-5-high | Anthropic | 1,660 | 12.6K |
| 9 |  Muse Spark 1.3 | Meta | 1,652 | 3.0K |
| 10 | qwen3.8-flash-next | Alibaba | 1,635 | 2.8K |
| 11 | Claude Fable 5 | Anthropic | 1,628 | 10.1K |
| 12 |  hy4-preview | Tencent | 1,624 | 2.3K |
| 13 |  Grok 4.6 | xAI | 1,618 | 4.3K |
| 14 | gpt-5.6-sol-xhigh (codex-harness) | OpenAI | 1,617 | 11.9K |
| 15 |  deepseek-v4.1-flash-max | DeepSeek | 1,614 | 1.4K |
| 16 |  GLM-5.3 | Zhipu AI | 1,614 | 3.7K |
| 17 | GGLM-5.3-Flash | Z.ai (Zhipu AI) | 1,607 | 2.9K |
| 18 | Qwen 3.8 27B | Alibaba | 1,593 | 4.9K |
| 19 | GLM-5.2 | Zhipu AI | 1,592 | 10.5K |
| 20 |  Gemini 3.7 Flash | 1,587 | 3.0K |
How this chart is built
Draws on a public blind human-preference arena, where scores come from real users voting in anonymous head-to-head comparisons. The chart shows its latest published results.
Positions are renumbered from high to low by this table's own metric, independent of any ranking the source publishes. Figures are the current real data with no manual adjustment; values we cannot obtain show as “—” and are never filled in with estimates.
FAQ
LMArena Code Leaderboard: who ranks #1?
In September 2026, OpenAI’s GPT-6 Astra ranks #1 (score: 1,800); claude-fable-5.1-max is second (1,758), followed by claude-opus-5-max (1,687).
LMArena Code Leaderboard: what are the top 10?
Ranked by score as of September 2026, the top 10 are: 1. GPT-6 Astra, 2. claude-fable-5.1-max, 3. claude-opus-5-max, 4. qwen3.8-max-0902, 5. Kimi K3, 6. kimi-k3-max, 7. Qwen 3.8 Max, 8. claude-opus-5-high, 9. Muse Spark 1.3, 10. qwen3.8-flash-next. The full top 20 is in the table on this page.
LMArena Code Leaderboard: where does the data come from, and how often is it updated?
Draws on a public blind human-preference arena, where scores come from real users voting in anonymous head-to-head comparisons. The chart shows its latest published results. Refreshed when the source publishes; the current data is dated September 15, 2026.