Best AI models for coding
Coding tests range from fresh competitive-programming problems to real GitHub issues and terminal tasks, where the model's patch or command sequence is executed and judged by tests, not by eye. Across 13 models with at least one coding score, Claude Opus 5.5 has the highest index at 97.0. The strongest open-weight model is DeepSeek V4 Pro (89.4). Among models you can deploy on OpenLLM Buddy, Qwen3.8 27B leads, ranked 7 overall.
Coding index
Average of each model's scores relative to the best published score on every test in this category (top = 100). Only tests a model reports are counted, and a model needs at least two of them to be ranked.
- OpenLLM Buddy catalog
- Open weights
- Closed API
- 1Claude Opus 5.5Anthropic · 2 of 5 tests97.0
- 2Claude Sonnet 5.5Anthropic · 2 of 5 tests95.2
- 3DeepSeek V4 ProDeepSeek · 4 of 5 tests89.4
- 4Reflection Beam 1.0Reflection AI · 4 of 5 tests89.1
- 5Claude Fable 5.1Anthropic · 2 of 5 tests84.7
- 6Tencent Hy4 PreviewTencent · 2 of 5 tests83.7
- 7Qwen3.8 27BAlibaba Qwen · 3 of 5 tests81.9
- 8Nemotron 3 UltraNVIDIA · 3 of 5 tests81.4
- 9GLM 5.3Z.ai (Zhipu) · 2 of 5 tests78.3
- 10Qwen3.6 27BAlibaba Qwen · 4 of 5 tests77.3
- 11MiMo V2.6 ProXiaomi · 2 of 5 tests74.3
- 12DeepSeek V4.1 FlashDeepSeek · 2 of 5 tests72.1
- 13Claude Haiku 5.5Anthropic · 2 of 5 tests63.8
16 more models report a single coding test and are listed in the table below without a rank: Kimi K3, GLM 5.3 Flash, Gemma 4 31B, Gemma 4 26B A4B, GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon, Gemini 3.8 Flash, Grok 4.7, Kimi K3.1 Pro, MiniMax M3.1 Flash, Qwen3.8 2.4T-A95B, Mistral Large 4, gpt-oss-120b, Muse Spark Open, Step 5 Preview.
Raw scores by test
Published percentages. A dot means the model has no published score on that test.
| Model | LCB | SWE-V | SWE-Pro | T-Bench | T-Bench 4 |
|---|---|---|---|---|---|
| Claude Opus 5.5 | · | · | 89.9 | · | 66.4 |
| Claude Sonnet 5.5 | · | · | 81.3 | · | 70.6 |
| DeepSeek V4 Pro | 93.5 | 80.6 | 55.4 | 87.9 | · |
| Reflection Beam 1.0 | 88.3 | 81.6 | 66.1 | 80.1 | · |
| Claude Fable 5.1 | · | · | 81.2 | · | 55.8 |
| Tencent Hy4 Preview | · | · | 65.7 | 85.4 | · |
| Qwen3.8 27B | 90.3 | · | 61.7 | 73.0 | · |
| Nemotron 3 Ultra | 89.0 | 70.7 | · | 56.4 | · |
| GLM 5.3 | · | · | · | 88.2 | 41.9 |
| Qwen3.6 27B | 83.9 | 77.2 | 53.5 | 59.3 | · |
| MiMo V2.6 Pro | · | · | · | 89.9 | 34.9 |
| DeepSeek V4.1 Flash | · | · | · | 90.6 | 31.2 |
| Claude Haiku 5.5 | · | · | 64.8 | · | 39.2 |
| Kimi K3 | · | · | · | 88.3 | · |
| GLM 5.3 Flash | · | · | · | 84.3 | · |
| Gemma 4 31B | 80.0 | · | · | · | · |
| Gemma 4 26B A4B | 77.1 | · | · | · | · |
| GPT-6 Astra | · | · | · | · | 59.1 |
| GPT-6.1 Sol | · | · | · | · | 56.1 |
| Gemini 4 Argon | · | · | · | · | 57.1 |
| Gemini 3.8 Flash | · | · | · | · | 19.7 |
| Grok 4.7 | · | · | · | · | 37.6 |
| Kimi K3.1 Pro | · | · | · | · | 65.7 |
| MiniMax M3.1 Flash | · | · | · | · | 38.1 |
| Qwen3.8 2.4T-A95B | · | · | · | 82.0 | · |
| Mistral Large 4 | · | · | · | · | 41.0 |
| gpt-oss-120b | · | 62.4 | · | · | · |
| Muse Spark Open | · | · | · | 76.4 | · |
| Step 5 Preview | · | · | · | · | 33.3 |
Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works