AI Coder Bench — Leaderboard

Models ranked18

Tasks1277

PeriodJune 2026

#	Model	Score	Aider ↗	SWE-bench ↗	LiveCode ↗	EvalPlus ↗
1↑6	Gemini 2.5 ProGoogle reasoningvisionlong-context	87.7	83.1	—	100.0	79.3
2↑4	Grok 4.3xAI reasoningagenticlong-context	86.0	79.6	—	100.0	80.5
3↑1	DeepSeek-V4-ProDeepSeek flagshipreasoningcoding	84.0	74.2	—	100.0	86.6
4↑8	DeepSeek-V4-FlashDeepSeek fastcheapcoding	83.5	—	—	—	83.5
5↑3	Claude Opus 4.8Anthropic flagshipreasoningvision	81.4	72.0	—	100.0	77.4
6↓4	GPT-5.4OpenAI workhorsecoding	77.7	61.7	—	100.0	89.0
7↓2	Claude Sonnet 4.6Anthropic workhorsecodingvision	75.6	64.9	70.6	100.0	81.7
8↑6	GPT-5.5OpenAI flagshipreasoning	74.7	88.0	65.0	100.0	11.0
9↓8	Qwen3.7-MaxQwen flagshipreasoningcoding	74.3	59.6	—	100.0	—
10↓1	Gemini 2.5 FlashGoogle workhorsevisionlong-context	72.1	55.1	—	100.0	75.6
11↑4	Mistral Large 3Mistral AI flagshipvisioncoding	62.2	—	—	—	62.2
12↓9	GPT-5.4 miniOpenAI fastcheap	61.3	32.9	59.8	100.0	89.0
13=	Kimi K2.6Moonshot AI flagshipreasoningvision	59.1	59.1	—	—	—
14↑2	Command R+ 08-2024Cohere flagshipreasoningcoding	56.7	—	—	—	56.7
15↓5	Claude Haiku 4.5Anthropic fastcheap	56.4	28.0	—	100.0	68.9
16↓5	Qwen3.6-PlusQwen codingreasoning	32.1	16.4	—	—	87.2
17↓4	CodestralMistral AI codingcode-completion	25.0	11.1	—	—	73.8
18=	Command R 03-2024Cohere codingmultilingualrag	12.0	12.0	—	—	—

S ≥ 90A ≥ 80B ≥ 70C < 70

Scores aggregated from public benchmarks · updated weekly

How scores are calculated

35%Aider Polyglot

225 real coding tasks across 6 languages

35%SWE-bench Verified

500 real GitHub issues, % resolved

20%LiveCodeBench

Ongoing competitive programming, Pass@1

10%EvalPlus (HumanEval+)

Stricter HumanEval test suite