Best LLMs for Coding

Gemini 3.1 Pro is the new top-ranked model on the Artificial Analysis Intelligence Index (score: 57) and leads on the coding benchmarks that matter most to developers. It scores 80.6% on SWE-Bench Verified (essentially tied with Claude Opus's 80.8%), 68.5% on Terminal-Bench 2.0, and 2887 Elo on LiveCodeBench Pro — far ahead of the competition on competitive and algorithmic coding. The dedicated gemini-3.1-pro-preview-customtools API endpoint is purpose-built for agentic pipelines that call bash, view_file, or search_code tools. At $2/$12 per 1M tokens (same as Gemini 3 Pro), the price-to-coding-capability ratio is unmatched at the frontier tier.

💰 API at $2/$12 per 1M tokens via Google AI Studio. Free developer tier available (rate-limited). Google One AI Premium ($20/month) for Gemini App access.

Claude Opus 4.6 is the most reliable coding model for complex, multi-step tasks — large refactors, implementing features across multiple files, debugging subtle logic errors. It tracks context across long conversations better than most models, which matters when a task spans hundreds of lines of code. SWE-Bench: 80.8% — a hair above Gemini 3.1 Pro. Still the best choice when you need a model that knows when to ask for clarification instead of guessing.

💰 API at $5/$25 per 1M tokens. Claude Max plan for consumer access ($100/month).

GPT-5.2 is the most tested coding LLM in the world and has the deepest ecosystem: GitHub Copilot, Cursor, and most IDE integrations are built around it. Strong on all standard coding tasks — generation, debugging, code explanation. The go-to choice if you want the most tools and integrations.

💰 Free tier at chatgpt.com. For IDE use, Cursor ($20/month) or GitHub Copilot ($10/month).

GPT-5 mini is the best budget coding option. Its reasoning model architecture handles multi-step problems — debugging, algorithm design, SQL — more reliably than non-reasoning models at the same price. At $0.25/$2.00 per 1M tokens, it's the smartest-per-dollar API choice for developers building coding assistants.

💰 Available on ChatGPT free tier with limits. API at $0.25/$2.00 per 1M tokens.

Claude Sonnet 4.6's 200K context window holds large codebases in context, and it's exceptional at following detailed code style and architecture instructions. Many developers prefer it for refactoring work and longer coding sessions where context retention matters more than raw speed.

💰 Free tier at claude.ai. API at $3/$15 per 1M tokens.

DeepSeek V3.2 is the sleeper pick for cost-sensitive coding API use. It matches GPT-5.2 on several coding benchmarks at a fraction of the price. If you're building AI-powered coding tools and can't use Chinese-hosted infrastructure for sensitive code, run it self-hosted.

💰 $0.27/$1.10 per 1M tokens via DeepSeek API. Cached input drops to $0.07/1M.

GPT OSS 120B is OpenAI's first open-weight model and the top-ranked open-weight reasoning model on Artificial Analysis (Intelligence Index 33). At $0.15/$0.60 per 1M tokens and 336 t/s, it's the fastest and cheapest path to a frontier-lab open model for coding pipelines. Download the weights from Hugging Face and self-host for near-zero cost. Best for developers who want OpenAI-quality coding in their own infrastructure.

💰 $0.15/$0.60 per 1M tokens via W&B Inference / OpenRouter. Self-host from Hugging Face for free (requires H100-class GPU).