TPSTokens Per Second

Best LLM for Coding in 2026

Ranked by composite coding score averaging 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
Claude Fable 5Anthropic
95.0%
avg
3
Claude Mythos PreviewAnthropic
93.9%
avg
4
GPT-5.6 LunaOpenAI
93.0%
avg
5
Claude Opus 4.8Anthropic
88.6%
avg
6
Claude Opus 4.7Anthropic
87.6%
avg
7
DeepSeek V4 ProDeepSeek
87.0%
avg
8
Grok 4.5xAI
86.6%
avg
9
Qwen3.7 MaxAlibaba / Qwen
86.0%
avg
10
Kimi K2.6Moonshot AI
84.9%
avg
11
Grok 4.20xAI
84.6%
avg
12
GPT-5.2OpenAI
84.5%
avg
13
GPT-5.4OpenAI
83.3%
avg
14
GLM-4.7Zhipu AI
81.6%
avg
15
Kimi K3Moonshot AI
80.7%
avg
16
MiniMax-M3MiniMax
80.5%
avg
17
MiniMax-M2.5MiniMax
80.2%
avg
18
Gemini 3.1 ProGoogle DeepMind
79.5%
avg
19
Gemini 3 FlashGoogle DeepMind
78.8%
avg
20
Qwen 3.6 PlusAlibaba / Qwen
78.8%
avg
21
Hy3Tencent
78.0%
avg
22
MiMo-V2-ProXiaomi
78.0%
avg
23
GLM 5Zhipu AI
77.8%
avg
24
Qwen3.7 PlusAlibaba / Qwen
77.7%
avg
25
Mistral Medium 3.5Mistral AI
77.6%
avg
26
Muse SparkMeta AI
77.4%
avg
27
Claude Opus 4.5Anthropic
77.3%
avg
28
Qwen3.6 27BAlibaba / Qwen
77.2%
avg
29
Qwen3.6-27B-FP8Alibaba / Qwen
77.2%
avg
30
Kimi K2.5Moonshot AI
76.8%
avg
31
Qwen3.5 397B A17BAlibaba / Qwen
76.4%
avg
32
MiniMax-M2MiniMax
76.0%
avg
33
Claude Opus 4.6Anthropic
76.0%
avg
34
Claude Sonnet 4.6Anthropic
75.2%
avg
35
MiMo-V2-OmniXiaomi
74.8%
avg
36
Laguna M.1Poolside
74.6%
avg
37
Step-3.5-FlashStepFun
74.4%
avg
38
Step-3.5-Flash-BaseStepFun
74.4%
avg
39
Hy3 PreviewTencent
74.4%
avg
40
MAI-Thinking-1Microsoft
73.5%
avg
41
Seed-2.0-LiteByteDance
73.5%
avg
42
Qwen3.6-35B-A3B-FP8Alibaba / Qwen
73.4%
avg
43
Qwen3.6 35B A3BAlibaba / Qwen
73.4%
avg
44
Claude Sonnet 4.5Anthropic
73.2%
avg
45
Qwen3-MaxAlibaba / Qwen
73.2%
avg
46
Claude Sonnet 5Anthropic
72.7%
avg
47
Qwen3.5-27BAlibaba / Qwen
72.4%
avg
48
Nemotron-3-Ultra-550B-A55BNVIDIA
71.9%
avg
49
Laguna XS 2.1Poolside
70.9%
avg
50
Qwen3.5-35B-A3BAlibaba / Qwen
69.2%
avg
51
GPT-5.4 miniOpenAI
68.5%
avg
52
Laguna XS.2Poolside
68.2%
avg
53
K-EXAONE 2.0 (750B-A37B)Lg Ai Research
68.2%
avg
54
North Mini Code 1.0Cohere
67.6%
avg
55
Grok 4 minixAI
64.8%
avg
56
Trinity Large ThinkingArcee Ai
63.2%
avg
57
GLM-5.2Zhipu AI
62.1%
avg
58
GLM-4.6Zhipu AI
62.0%
avg
59
Devstral MediumMistral AI
61.6%
avg
60
NVIDIA Nemotron-3-Super-120B-A12BNVIDIA
60.5%
avg
61
MiMo-V2.5-ProXiaomi
59.3%
avg
62
Claude Haiku 4.5Anthropic
58.0%
avg
63
MiMo-V2.5Xiaomi
56.1%
avg

How we rank coding models

We average three industry-standard benchmarks:

  • 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.