TPSTokens Per Second

Best LLM for Coding in 2026

Ranked by composite coding score averaging HumanEval, SWE-bench Verified, and LiveCodeBench. All scores from official model cards and technical reports.

Updated automatically as new models release. Full benchmark leaderboard →

1
GPT-5.6 SolOpenAI
96.2%
avg
2
DiffusionGemma 26B A4B IT NVFP4NVIDIA
95.0%
avg
3
Claude Fable 5Anthropic
95.0%
avg
4
Claude Mythos PreviewAnthropic
93.9%
avg
5
GPT-5.6 LunaOpenAI
93.0%
avg
6
Qwen3 72BAlibaba / Qwen
91.2%
avg
7
o1-proOpenAI
91.0%
avg
8
Grok 4.20xAI
89.1%
avg
9
Claude Opus 4.8Anthropic
88.6%
avg
10
Granite 4.1 30BIbm
88.4%
avg
11
GPT-5.4OpenAI
88.1%
avg
12
gpt-oss-120bOpenAI
87.8%
avg
13
Claude Opus 4.7Anthropic
87.6%
avg
14
Grok 4.5xAI
86.6%
avg
15
Qwen3.7 MaxAlibaba / Qwen
86.0%
avg
16
Grok 4xAI
85.7%
avg
17
Granite 4.1 8BIbm
85.4%
avg
18
Kimi K2 ThinkingMoonshot AI
85.3%
avg
19
GPT-5.4 NanoOpenAI
85.2%
avg
20
Gemini 3.1 ProGoogle DeepMind
84.9%
avg
21
Kimi K2.6Moonshot AI
84.9%
avg
22
o3OpenAI
84.7%
avg
23
GPT-5.2OpenAI
84.5%
avg
24
DeepSeek-V4-ProDeepSeek
83.6%
avg
25
Sonar Reasoning ProPerplexity AI
82.0%
avg
26
GLM-4.7Zhipu AI
81.6%
avg
27
Claude Opus 4.6Anthropic
81.3%
avg
28
GPT-5.4 miniOpenAI
81.0%
avg
29
Trinity Large ThinkingArcee Ai
80.7%
avg
30
Kimi K3Moonshot AI
80.7%
avg
31
DeepSeek-V4-FlashDeepSeek
80.5%
avg
32
Claude Sonnet 4.6Anthropic
80.5%
avg
33
Sonar Deep ResearchPerplexity AI
80.5%
avg
34
MiniMax M3MiniMax
80.5%
avg
35
MiniMax-M2.5MiniMax
80.2%
avg
36
Gemma 4 31BGoogle DeepMind
80.0%
avg
37
NVIDIA Nemotron-Labs-3-Puzzle-75B-A9B-NVFP4NVIDIA
79.9%
avg
38
QwQ-32BAlibaba / Qwen
79.3%
avg
39
GPT-4.5OpenAI
79.0%
avg
40
Claude Sonnet 4.5Anthropic
78.9%
avg
41
Gemini 3 FlashGoogle DeepMind
78.8%
avg
42
Qwen 3.6 PlusAlibaba / Qwen
78.8%
avg
43
Gemini 2.5 ProGoogle DeepMind
78.4%
avg
44
Claude Opus 4Anthropic
78.3%
avg
45
Sonar ProPerplexity AI
78.2%
avg
46
Hy3Tencent
78.0%
avg
47
MiMo-V2-ProXiaomi
78.0%
avg
48
GLM 5Zhipu AI
77.8%
avg
49
Kimi K2 0905Moonshot AI
77.8%
avg
50
Step-3.5-Flash-BaseStepFun
77.8%
avg
51
gpt-oss-20bOpenAI
77.7%
avg
52
Qwen3.7 PlusAlibaba / Qwen
77.7%
avg
53
Mistral Medium 3.5Mistral AI
77.6%
avg
54
Muse SparkMeta AI
77.4%
avg
55
Claude Opus 4.5Anthropic
77.3%
avg
56
Qwen3.6 27BAlibaba / Qwen
77.2%
avg
57
Qwen3.6-27B-FP8Alibaba / Qwen
77.2%
avg
58
Gemma 4 26B A4BGoogle DeepMind
77.1%
avg
59
Sonar ReasoningPerplexity AI
77.0%
avg
60
o4-miniOpenAI
76.9%
avg
61
Kimi K2.5Moonshot AI
76.8%
avg
62
Qwen3.5 397B A17BAlibaba / Qwen
76.4%
avg
63
MiniMax-M2MiniMax
76.0%
avg
64
Nemotron-Labs-TwoTower-30B-A3B-Base-BF16NVIDIA
75.6%
avg
65
MiMo-V2-OmniXiaomi
74.8%
avg
66
Claude 3.7 SonnetAnthropic
74.8%
avg
67
GPT-4.1OpenAI
74.7%
avg
68
Laguna M.1Poolside
74.6%
avg
69
Step-3.5-FlashStepFun
74.4%
avg
70
Hy3 PreviewTencent
74.4%
avg
71
DeepSeek R1DeepSeek
74.1%
avg
72
Grok 4 minixAI
73.8%
avg
73
Seed-2.0-LiteByteDance
73.5%
avg
74
MAI-Thinking-1Microsoft
73.5%
avg
75
Qwen3.6-35B-A3B-FP8Alibaba / Qwen
73.4%
avg
76
Qwen3.6 35B A3BAlibaba / Qwen
73.4%
avg
77
Qwen3-MaxAlibaba / Qwen
73.2%
avg
78
Claude Sonnet 5Anthropic
72.7%
avg
79
SonarPerplexity AI
72.5%
avg
80
Qwen3.5-27BAlibaba / Qwen
72.4%
avg
81
Gemma 4 12B UnifiedGoogle DeepMind
72.0%
avg
82
Nemotron-3-Ultra-550B-A55BNVIDIA
71.9%
avg
83
Laguna XS 2.1Poolside
70.9%
avg
84
Gemma 4 31B IT NVFP4NVIDIA
70.6%
avg
85
Mellum2-12B-A2.5B-ThinkingJetBrains
69.9%
avg
86
Llama 4 MaverickMeta AI
69.8%
avg
87
Claude Haiku 4.5Anthropic
69.8%
avg
88
Qwen3.5-35B-A3BAlibaba / Qwen
69.2%
avg
89
Laguna XS.2Poolside
68.2%
avg
90
North Mini Code 1.0Cohere
67.6%
avg
91
Mercury 2Inception
67.0%
avg
92
Grok 3xAI
66.4%
avg
93
ZAYA1-8BZyphra
65.8%
avg
94
MiMo-V2.5-ProXiaomi
64.7%
avg
95
GLM-5.2Zhipu AI
62.1%
avg
96
GLM-4.6Zhipu AI
62.0%
avg
97
Mistral Medium 3Mistral AI
62.0%
avg
98
Devstral MediumMistral AI
61.6%
avg
99
GPT-4oOpenAI
60.7%
avg
100
NVIDIA Nemotron-3-Super-120B-A12BNVIDIA
60.5%
avg
101
MiMo-V2.5Xiaomi
56.1%
avg
102
Gemini 3.5 FlashGoogle DeepMind
56.0%
avg
103
Magistral SmallMistral AI
51.4%
avg
104
Mistral Large 3Mistral AI
46.5%
avg
105
Devstral 2Mistral AI
44.8%
avg
106
Gemma 4 E2BGoogle DeepMind
44.0%
avg
107
Mistral Small 4Mistral AI
0.7%
avg

How we rank coding models

We average three industry-standard benchmarks:

  • HumanEval — Function-completion coding tasks in Python. Tests whether a model can write correct, working code from a docstring description. Pass@1 accuracy.
  • SWE-bench Verified — Real GitHub issues from popular open-source repos. Tests autonomous software engineering: read the issue, write a fix, pass the test suite. The most practical real-world coding benchmark available.
  • LiveCodeBench — Competitive programming problems from LeetCode, Codeforces, and AtCoder, collected after model training cutoffs to prevent contamination. Harder than HumanEval.

All scores are from official model cards, technical reports, or the HuggingFace Open LLM Leaderboard. Rankings update automatically as new models are released.

Also see: Best Reasoning LLM, Best Cheap LLM, Compare any two models.