Best AI for coding
Writing new code, debugging, refactoring, and reviewing pull requests.
Coding has become the single most important task for production AI. Agents commit code. IDEs rely on models for tab completion, multi-file edits, and test generation. The cost of a wrong answer compounds — a model that hallucinates an API gets wedged into a codebase and quietly ships.
What matters: correctness on held-out tests (SWE-bench, LiveCodeBench), diff-style edit fidelity, tool-use fluency (shell, file I/O, search), and the ability to reason about a large context without losing the plot. Speed matters too — autocomplete needs to fire in under 400ms to feel natural.
Our coding rank weights SWE-bench Verified (60%), multi-file edit accuracy (25%), and tool-use reliability (15%). If you're building an autonomous coding agent, weight reasoning and tool-use more heavily. For interactive autocomplete, speed becomes the deciding factor.
Full rankings
All 22 models, scored for coding
| # | Model | Provider | Task score | $/in | $/out | Context |
|---|---|---|---|---|---|---|
| 01 | Anthropic: Claude Fable 5 | AAnthropic | 95.0% | $10.00 | $50.00 | 1M |
| 02 | Claude Opus 4.7 | AAnthropic | 83.5% | $5.00 | $25.00 | 1M |
| 03 | GPT-5.5 | OOpenAI | 80.6% | $5.00 | $30.00 | 1.1M |
| 04 | Gemini 3.5 Flash | GGoogle | 79.3% | $1.50 | $9.00 | 1.0M |
| 05 | Z.ai: GLM 5.2 | ZZhipu AI | 78.7% | $0.406 | $1.28 | 1.0M |
| 06 | Claude Opus 4.6 | AAnthropic | 78.7% | $5.00 | $25.00 | 1M |
| 07 | DeepSeek V4 Pro | DDeepSeek | 77.6% | $0.435 | $0.870 | 1.0M |
| 08 | Qwen 3.7 Max | QQwen | 77.3% | $1.25 | $3.75 | 1M |
| 09 | GPT-5.4 | OOpenAI | 76.9% | $2.50 | $15.00 | 1.1M |
| 10 | MoonshotAI: Kimi K2.6 | MMoonshotAI | 76.7% | $0.660 | $3.41 | 262K |
| 11 | Qwen: Qwen3.6 Max Preview | QQwen | 76.7% | $1.04 | $6.24 | 262K |
| 12 | Gemini 3 Flash Preview | GGoogle | 75.4% | $0.500 | $3.00 | 1.0M |
| 13 | Claude Sonnet 4.6 | AAnthropic | 75.2% | $3.00 | $15.00 | 1M |
| 14 | GPT-5.2 | OOpenAI | 73.8% | $1.75 | $14.00 | 400K |
| 15 | GPT-5 | OOpenAI | 73.6% | $1.25 | $10.00 | 400K |
| 16 | GPT-5.1 | OOpenAI | 68.0% | $1.25 | $10.00 | 400K |
| 17 | GPT-5 Mini | OOpenAI | 64.7% | $0.250 | $2.00 | 400K |
| 18 | o3 | OOpenAI | 62.3% | $2.00 | $8.00 | 200K |
| 19 | Qwen: Qwen3.6 Plus | QQwen | 57.9% | $0.325 | $1.95 | 1M |
| 20 | Gemini 2.5 Pro | GGoogle | 57.6% | $1.25 | $10.00 | 1.0M |
| 21 | GPT-4.1 | OOpenAI | 48.5% | $2.00 | $8.00 | 1.0M |
| 22 | GPT-4o | OOpenAI | 31.0% | $2.50 | $10.00 | 128K |
Pricing — top 5 for coding
The best AI for coding changes every month.
We'll email you when rankings shift, new models hit the top 5, or pricing cuts reshuffle the value leaders.