Checked against the OpenRouter public models feed. Prices are copied, not estimated.

The Best Model ForWhich AI model to use, from ten labs.

The best AI model for coding

Build a feature, fix a bug, or draft a script you will run.

Short answer

Start with Claude Sonnet 5.5 ($2.00 in · $10.00 out per million tokens). Sonnet 5.5 is the reviewed default for a well-scoped feature or bug. The listing aims it at building features and fixing bugs, and the window is large enough for a spec plus the files you paste. A cheaper stop is Claude Haiku 5.5 ($0.10 in · $0.50 out).

Quality

Claude Sonnet 5.5

Anthropic · $2.00 in · $10.00 out

Sonnet 5.5 is the reviewed default for a well-scoped feature or bug. The listing aims it at building features and fixing bugs, and the window is large enough for a spec plus the files you paste.

Claude Sonnet 5.5 is listed at $2.00 input and $10.00 output per million tokens, with a 1,000,000-token context window.

Balanced

GPT-6.2 Sol

OpenAI · $2.00 in · $10.00 out

Sol is the everyday OpenAI model under the Astra flagship. Use it for a small web app when you also want to attach a screenshot.

GPT-6.2 Sol is listed at $2.00 input and $10.00 output per million tokens, with a 1,050,000-token context window.

Cheapest sensible

DeepSeek V4.1 Flash 1017

DeepSeek · $0.15 in · $0.60 out

V4.1 Flash is open weights, accepts a screenshot, and is listed like a budget model. Start here when the coding bill matters. Move up only if your tests say the answer is shallow.

DeepSeek V4.1 Flash 1017 is listed at $0.15 input and $0.60 output per million tokens, with a 1,048,576-token context window.

Also on the feed

These match the factual limits of the task (modality, tools, context) and are not the reviewed picks. A row marked new has no guide note.

Questions

Should I use a flagship for a small web app?

Usually no. A small app with a written spec is a Sonnet, Sol, or Flash job. Keep Astra, Fable, or Grok 4.7 for a design you cannot afford to redo.

Which coding model can I host?

The reviewed open-weight coding options are Devstral 2, Qwen3 Coder Next, and the DeepSeek V4 rows. Open weights means the feed lists a Hugging Face id. It does not mean hosting is free.