The best coding model right now
Scored on what developers actually run inside their coding tools, re-counted every night.
45 models ranked · 36 coding tools tracked · captured 3 August 2026 · movement against 2 August 2026
| Rank | Model | Per million | Context | Routed through | Volume | Movement |
|---|---|---|---|---|---|---|
| 1 | Anthropic: Claude Opus 4.8anthropic/claude-opus-4.8 | $10.00 | 1M | 862B | Held | |
| 2 | MoonshotAI: Kimi K3moonshotai/kimi-k3 | $6.00 | 1.0M | 257B | Held | |
| 3 | Anthropic: Claude Sonnet 4.6anthropic/claude-sonnet-4.6 | $6.00 | 1M | 276B | Held | |
| 4 | Claude Opus 5anthropic/claude-opus-5 | $10.00 | 1M | 290B | 1 | |
| 5 | Anthropic: Claude Fable 5anthropic/claude-fable-5 | $20.00 | 1M | 388B | 1 | |
| 6 | Anthropic: Claude Opus 4.6anthropic/claude-opus-4.6 | $10.00 | 1M | 108B | Held | |
| 7 | OpenAI: GPT-5.6 Solopenai/gpt-5.6-sol | $11.25 | 1.1M | 242B | Held | |
| 8 | Anthropic: Claude Opus 4.7anthropic/claude-opus-4.7 | $10.00 | 1M | 111B | Held | |
| 9 | OpenAI: GPT-5.6 Sol Proopenai/gpt-5.6-sol-pro | $11.25 | 1.1M | 91B | Held |
The tracker, in numbers
2026-08-03Coding tools tracked
36
2026-08-03IDEs, CLI agents, cloud agents, extensions, app builders and design-to-code — tracked whether or not they carry measurable traffic.Models in the eligible set
339
2026-08-03Text input, tool-calling support and at least a 32K context window. Everything else is excluded before ranking.Tools that are not alive
7
2026-08-033 acquired · 2 sunset · 1 stalled · 1 archived. Each one is still being recommended somewhere as a live product.Ranked with no published price
1
2026-08-03Shown as not published, never as $0. A model published at no cost and a model with no published cost are different facts.Nobody else re-reads the pricing pages
Every listicle ranking AI coding tools was written once and has been decaying ever since. A free tier is withdrawn, a company is acquired, a repository is archived — and the article recommending it stays exactly where it was, at the top of the results. This site reads each tracked tool’s own pricing page, its home page and its public repository once a night and publishes what came back different, with a link to the evidence on every entry.
7 of the 36 tools tracked here are no longer what they were, and they stay listed on purpose — a tool that was acquired or archived does not disappear from the write-ups already published about it. The board is the same discipline pointed at models. It ranks adoption — the tokens developers put through a model inside these tools — and it runs no benchmark at all. Rank 1 is the model being used the most, not the one that writes the best code, and the method page says so before it says anything else.
Last run finished 19 hours ago.
