最佳AI编程模型

基于编程基准测试、价格和实际性能的客观排名。每周更新。

已排名模型数

Claude Fable 5

顶级模型

100

最高分数

编程排行榜

#	模型	分数	价格（1M令牌）	上下文	最适合
1	Claude Fable 5 anthropic	100	$10 / $50 每1M	1M	Complex architecture, Multi-file refactoring
2	Kimi K3 moonshotai	99.62	$3 / $15 每1M	1M	Complex reasoning and architecture, Multi-language code generation
3	GPT-5.6 terra openai	99.5	$10 / $60 每1M	1M	Autonomous agentic workflows, Large-scale refactoring
4	GPT-5.6 sol openai	98.8	$5 / $30 每1M	1M	GitHub Copilot, Algorithm design
5	GPT-5.6 Luna openai	98.1	$1.5 / $9 每1M	400K	Real-time completions, High-volume inline suggestions
6	GPT-5.5 openai	97.91	$5 / $30 每1M	1M	GitHub Copilot, Algorithm design
7	Claude Opus 4.8 anthropic	97.07	$5 / $25 每1M	1M	Complex architecture, Multi-file refactoring
8	Qwen3.8 Max Preview qwen	95	$2.5 / $10 每1M	128K	Full-stack development, Multilingual coding projects
9	Grok 4.5 x-ai	94.65	$2 / $6 每1M	500K	Full-stack development, Agentic coding workflows
10	GLM-5.2 z-ai	89.89	$0.93 / $3 每1M	1M	Chinese market, Bilingual development
11	DeepSeek V4 Flash (0731) deepseek	88.12	$0.09 / $0.18 每1M	1M	Algorithm design, Full-stack development
12	Qwen3.7 Max qwen	86.24	$1.25 / $3.75 每1M	1M	Open-source projects, Code generation
13	Claude Sonnet 4.6 anthropic	82.4	$3 / $15 每1M	1M	Claude Code, Code review
14	Kimi K2.7 Code moonshotai	79.44	$0.74 / $3.5 每1M	262K	Code generation, Bilingual coding
15	MiMo V2.5 Pro mimo	78.69	$0.44 / $0.87 每1M	1M	Coding tasks, Reasoning
16	DeepSeek V4 Pro deepseek	77.61	$0.44 / $0.87 每1M	1M	Algorithm design, Full-stack development
17	Muse Spark meta	76.63	$0 / $0 每1M	128K	Self-hosting, Research
18	MiniMax M3 minimax	76.57	$0.3 / $1.2 每1M	1M	Full-stack development, Bilingual coding
19	HY3 tencent	73.5	$0.13 / $0.53 每1M	262K	Bilingual development (CN/EN), Budget reasoning tasks
20	GPT-5.4 mini openai	73.31	$0.75 / $4.5 每1M	400K	GitHub Copilot, Algorithm design
21	GLM-5.1 z-ai	72.93	$0.98 / $4.3 每1M	203K	Chinese market, Bilingual development
22	Kimi K2.6 moonshotai	72.5	$0.55 / $3.2 每1M	262K	Web development, Bilingual coding
23	Gemini 3.5 Flash google	70	$1.5 / $9 每1M	1M	Large codebase analysis, Documentation generation
24	Gemini 3.1 Pro google	65	$2 / $12 每1M	1M	Large codebase analysis, Documentation generation
25	Nemotron 3 Ultra nvidia	64.41	$0.5 / $2.2 每1M	1M	NVIDIA ecosystem, Hardware-optimized tasks
26	Qwen3.5 Coder qwen	63.03	$0.11 / $0.8 每1M	262K	Open-source projects, Code generation
27	Mistral Medium 3.5 mistralai	61.32	$1.5 / $7.5 每1M	262K	European market, Code generation
28	Claude Haiku 4.5 anthropic	57.38	$1 / $5 每1M	200K	Quick completions, Simple functions
29	Gemma 4 31B google	56.78	$0 / $0 每1M	262K	Self-hosting, Research
30	Grok 4.3 x-ai	55.23	$1.25 / $2.5 每1M	1M	Reasoning tasks, Creative coding

方法论

我们的编程分数综合了多个基准测试，包括HumanEval、MBPP、SWE-bench和LiveCodeBench。分数根据基准测试与现实世界编程任务的相关性进行加权。价格反映截至2026年6月的标准API费率。