← the full leaderboard · coding
The best AI model for coding, ranked.
Claude Fable 5 leads the coding ranking today, ahead of Claude Opus 5.5 and Claude Fable 5.1. LMArena · Coding, LiveBench · Coding and Vals Vibe Code Bench: 31 models combined into one coding index, updated daily. Free and public.
Scores as published by each board; table composed September 28, 2026.
Get the next launch alert, free →
Coding leaderboard
| # | Model | Coding index | LMArena · Coding | LiveBench · Coding | Vals Vibe Code Bench | API price / 1M |
|---|---|---|---|---|---|---|
| 1 | Claude Fable 5 · Anthropic | 100 | 1,551 | 74.1 | 90.4 | $10 / $50 |
| 2 | Claude Opus 5.5 · Anthropic | 96.7 | 1,547 | 80.5 | 90.3 | $4 / $20 |
| 3 | Claude Fable 5.1 · Anthropic | 93.3 | 1,529 | 76.2 | 90.3 | $10 / $50 |
| 4 | GPT-6 Astra · OpenAI | 90 | 1,542 | 68.8 | 89.6 | $10 / $50 |
| 5 | Claude Opus 5 · Anthropic | 86.7 | 1,533 | 73.3 | 88.4 | $5 / $25 |
| 6 | Muse Spark 1.3 · Meta | 83.3 | 1,539 | 72.6 | 85.9 | $1.25 / $4.25 |
| 7 | Kimi K3 · Moonshot | 80 | 1,541 | 71.8 | 85.0 | $3 / $15 |
| 8 | DeepSeek V4.1 Flash · DeepSeek | 72.4 | 1,532 | 78.7 | 84.7 | — |
| 9 | Claude Opus 4.8 · Anthropic | 66.7 | 1,534 | 66.2 | 82.7 | $5 / $25 |
| 10 | GPT-5.6 · OpenAI | 65.5 | 1,531 | 70.1 | 80.5 | $4 / $20 |
| 11 | Claude Sonnet 5 · Anthropic | 56.7 | 1,519 | 70.0 | 81.3 | $2 / $10 |
| 12 | GPT-6 Sol · OpenAI | 55.2 | 1,524 | 67.3 | 87.8 | $2 / $10 |
| 13 | Muse Spark 1.1 · Meta | 53.3 | 1,532 | 67.8 | 72.2 | $1.25 / $4.25 |
| 14 | GLM-5.3 Flash · Zhipu AI | 51.7 | 1,523 | 67.9 | 30.8 | $0.15 / $0.50 |
| 15 | Gemini 3.8 Flash · Google | 50 | 1,532 | 63.4 | 78.7 | $0.75 / $3.75 |
| 16 | GLM-5.3 · Zhipu AI | 48.3 | 1,522 | 69.9 | 78.1 | $1.40 / $4.40 |
| 17 | Grok 4.6 · xAI | 43.3 | 1,507 | 66.9 | 76.2 | $2 / $6 |
| 18 | Qwen 3.8 Max · Alibaba | 41.4 | 1,520 | 68.8 | 64.7 | $2 / $6 |
| 19 | DeepSeek V4 Pro · DeepSeek | 40 | 1,506 | 66.1 | 82.3 | $0.435 / $0.87 |
| 20 | GPT-5.5 · OpenAI | 36.7 | 1,519 | 68.1 | 69.8 | $5 / $30 |
| 21 | Grok 4.7 · xAI | 33.3 | 1,488 | 65.6 | 86.2 | $2 / $6 |
| 22 | GPT-6 Luna · OpenAI | 30 | 1,515 | 65.1 | 81.6 | $0.10 / $0.50 |
| 23 | GLM-5.2 · Zhipu AI | 26.7 | 1,512 | 65.7 | 64.0 | $1.40 / $4.40 |
| 24 | Grok 4.5 · xAI | 24.1 | 1,513 | 62.5 | 69.0 | $2 / $6 |
| 25 | Gemini 3.5 Flash · Google | 23.3 | 1,507 | 63.6 | 48.7 | $1.50 / $9 |
| 26 | Kimi K2.6 · Moonshot | 20 | 1,516 | 62.7 | 37.9 | — |
| 26 | Qwen 3.7 Max · Alibaba | 20 | 1,525 | 58.9 | 47.7 | — |
| 28 | Gemini 3.1 Pro · Google | 13.3 | 1,521 | 60.3 | 32.0 | — |
| 29 | Kimi K2.7 Code · Moonshot | 10 | — | 59.8 | 47.2 | — |
| 30 | MiniMax M3 · MiniMax | 6.9 | 1,495 | 54.4 | 47.6 | $0.30 / $1.20 |
| 31 | Inkling · Thinking Machines | 3.4 | 1,491 | 60.2 | 19.2 | — |
Coding index: the median of each model's percentile across the boards that score it (at least two). Prices are input / output per 1M tokens on the maker's own API. A dash means that board doesn't score the model, or the maker publishes no API price.
Quick answers
What is the best AI model for coding right now?
Claude Fable 5 at the top of the ArtificialWatch coding index (100), ahead of Claude Opus 5.5 (96.7) and Claude Fable 5.1 (93.3). The index is the median of each model's percentile across LMArena · Coding, LiveBench · Coding and Vals Vibe Code Bench, so no single board decides it; each board's own score is in the table.
What is the best open-weights model for coding?
Kimi K3 from Moonshot, the highest-ranked model with open weights on the coding index (80, #7 overall).
What is the cheapest top-10 AI model for coding?
Among the ten highest-ranked models with a price on their maker's own API, Muse Spark 1.3 (#6) has the lowest output price: $1.25 input / $4.25 output per 1M tokens.
Which benchmarks does the coding ranking combine?
LMArena · Coding (Elo (community votes on coding prompts)); LiveBench · Coding (mean of its Coding and Agentic Coding categories (0-100)); Vals Vibe Code Bench (accuracy %). Each model's best-variant score on each board becomes a percentile within that board, and a model needs at least 2 boards to get an index score. Vals' SWE-bench Verified, LiveCodeBench and Terminal-Bench are left out: Vals archived them and no longer runs new models on them.
Sources
- LMArena · Coding · Elo (community votes on coding prompts) · 30 models · read September 28, 2026
- LiveBench · Coding · mean of its Coding and Agentic Coding categories (0-100) · 31 models · read September 28, 2026
- Vals Vibe Code Bench · accuracy % · 31 models · read September 28, 2026
The full LLM leaderboard · Open-weights models ranked · Every model's API price · Head-to-head comparisons
Hear the minute the next one drops. Free.
A free Chrome push the moment a new model answers on a public API, an email about 15 minutes later. Texts or a phone call on paid plans.
- Free forever
- No card
- Unsubscribe in one click