Rankings

What people actually run

Leaderboards by tokens processed through the gateway, rolled up weekly from the event stream — one immutable event per request, aggregated after the fact.

Top models Sample data

Week of 20 Jul 2026 · 13.45B tokens · 21.6M requests

#ΔModelTokensRequestsAppsIn / 1M
1Claude Sonnet 5anthropic/claude-sonnet-54.82B6.1M1240$3
2▲ 1Claude Opus 5anthropic/claude-opus-53.15B2.4M980$5
3▼ 1Claude Haiku 4.5anthropic/claude-haiku-4-52.44B9.7M860$1
4Claude Sonnet 4.6anthropic/claude-sonnet-4-61.18B1.9M410$3
5Claude Opus 4.8anthropic/claude-opus-4-8742.0M610.0K320$5
6Claude Fable 5anthropic/claude-fable-5486.0M210.0K145$10
7Claude Opus 4.7anthropic/claude-opus-4-7268.0M240.0K130$5
8Claude Opus 4.6anthropic/claude-opus-4-6154.0M160.0K95$5
9Claude Sonnet 4.5anthropic/claude-sonnet-4-5121.0M190.0K88$3
10Claude Opus 4.5anthropic/claude-opus-4-567.0M72.0K51$5
11Claude Opus 4.1anthropic/claude-opus-4-121.0M26.0K19$15

Trailing 4 weeks Sample data

29 Jun 2026 – 20 Jul 2026 · 50.27B tokens

Market share Sample data

Token share by model series · week of 20 Jul 2026

  • sonnet45.5%6.12B
  • opus32.7%4.40B
  • haiku18.1%2.44B
  • fable3.6%486.0M

Top models by task Sample data

Traffic split by request shape · week of 20 Jul 2026

Coding

Code generation, review, refactors and repository work.

  1. 1Claude Sonnet 5
  2. 2Claude Opus 5
  3. 3Claude Sonnet 4.6
  4. 4Claude Opus 4.8
  5. 5Claude Fable 5

Agents

Multi-step tool-using runs and long-horizon autonomy.

  1. 1Claude Sonnet 5
  2. 2Claude Opus 5
  3. 3Claude Haiku 4.5
  4. 4Claude Opus 4.8
  5. 5Claude Sonnet 4.6

Extraction

Classification, structured extraction and routing.

  1. 1Claude Haiku 4.5
  2. 2Claude Sonnet 5
  3. 3Claude Sonnet 4.6
  4. 4Claude Opus 5
  5. 5Claude Opus 4.8

Writing

Drafting, summarisation and long-form prose.

  1. 1Claude Sonnet 5
  2. 2Claude Opus 5
  3. 3Claude Haiku 4.5
  4. 4Claude Sonnet 4.6
  5. 5Claude Opus 4.8