Skip to content

Which coding model is fastest right now?

Live response speed from Claude Code, Codex, Grok Build, etc. Privacy-safe open source clients. Free, no account needed.

Output tokens per second while the model is responding, excluding tool runs and your time.

Latest community upload ·

Completed turns · uploads usually take 1–7 min

Leaderboard

Ranked by median · last 24 hours
  1. Response speed · last 24 hours
    0200
    125tok/s
    While the model responds. Excludes tools and your time.

    AnthropicAclaude-sonnet-5-5Subagent

    Anthropic · Claude Code

    Effortmedium
    Slow end (p10)
    113 tok/s
    Slowest tenth of contributors
    Observed range
    61.5–157 tok/s
    Contributors · responses
    10+ · 1k+
    30+ turns
    Last observed
    1 h ago

    By coding tool and effort

    CohortResponse speedTurn speedCoverage
    Claude Code 2.1.321SubagentEffort: medium125 tok/s10+ contributors · 1k+ responses64.5 tok/sIncludes tools and waiting10+ contributors10+ turns
    Claude Code 2.1.317SubagentEffort: medium119 tok/sa few contributors · 100+ responses55.8 tok/sIncludes tools and waitinga few contributors10+ turns
    Claude Code 2.1.325SubagentEffort: medium95.4 tok/sa few contributors · 100+ responses37.2 tok/sIncludes tools and waitinga few contributorsa few turns
    Claude Code 2.1.326SubagentEffort: medium155 tok/sa few contributors · 30+ responses69.9 tok/sIncludes tools and waitinga few contributorsa few turns
    Claude Code 2.1.323SubagentEffort: medium124 tok/sa few contributors · 100+ responses24.3 tok/sIncludes tools and waitinga few contributorsa few turns

    History

    010020006:55 PM05:55 AM05:55 PM

    16 published time buckets. Gaps mean missing or insufficient observations.

    Now vs previous period

    Last 15 minutes
    — tok/s
    Change
    4.2% faster
    previous 24 hours: 120 tok/s
    Evidence and rulesRecent buckets, baseline and limits
    Recent · last 15 minutes

    Unavailable

    Insufficient observations

    Completed five-minute observation buckets. A fresh page does not imply fresh measurements.

    Last 24 hours compared with the previous period
    Previous equal period120 tok/s30+ turns · a few contributors
    Change in response speed4.2% higherHigher means more output per second of response

    Different tasks and contributors can explain changes. This comparison is separate from slowdown detection.

    Geographic coverage is unknown. These reports cannot establish worldwide provider health. Output quality, price and Fast mode are not measured.

Coverage uses approximate bands to protect contributor privacy: a few means 1–9; higher counts show a lower bound, such as 10+ or 1k+. Coverage bands.

Detailed measurementsCoverage, ranges and percentiles
Medians and percentiles weight each contributor equally; min/max are the 5th and 95th percentiles of eligible individual turns in the selected window, weighting each contributor equally. Coding tools, measurement definitions, inference providers, models, recorded effort and tool versions stay separate. Rates from different measurement definitions are not directly comparable. Response speed counts only the time the model is responding (first-token wait included, tools and your time excluded); turn speed includes tools and waiting. Missing effort remains unknown. Workload and speed tier are uncontrolled; this is neither generation speed nor an answer-quality ranking.
Model / providerResponse speedTurn speedFirst tokenCoverage
claude-fable-5-2Anthropic · Claude Code 2.1.321Effort: highclaude-transcript-v4 · claude-observed-turn-v185.2 tok/sMin 59.1 tok/s · Max 109 tok/sSlow-end p10: 74.8 tok/s10+ contributors · 100+ turns · 100+ responses64.5 tok/sMin 35.5 tok/s · Max 103 tok/sSlow-end p10: 52 tok/s10+ contributors · 100+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors100+ turns
claude-fable-5-2Anthropic · Claude Code 2.1.321Effort: mediumclaude-transcript-v4 · claude-observed-turn-v186.2 tok/sMin 66.5 tok/s · Max 107 tok/sSlow-end p10: 76.5 tok/s10+ contributors · 30+ turns · 100+ responses39.3 tok/sMin 4.5 tok/s · Max 102 tok/sSlow-end p10: 12.9 tok/s10+ contributors · 30+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors30+ turns
claude-fable-5-2Anthropic · Claude Code 2.1.323Effort: highclaude-transcript-v4 · claude-observed-subagent-turn-v180.4 tok/sMin 72.6 tok/s · Max 92.5 tok/sSlow-end p10: 74.1 tok/sa few contributors · a few turns · 10+ responses64.2 tok/sMin 57.9 tok/s · Max 73.8 tok/sSlow-end p10: 59.2 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-fable-5-2Anthropic · Claude Code 2.1.323Effort: highclaude-transcript-v4 · claude-observed-turn-v183.8 tok/sMin 66.5 tok/s · Max 102 tok/sSlow-end p10: 70.4 tok/s10+ contributors · 10+ turns · 100+ responses41.3 tok/sMin 5.8 tok/s · Max 82.6 tok/sSlow-end p10: 9.1 tok/s10+ contributors · 10+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors10+ turns
claude-fable-5-2Anthropic · Claude Code 2.1.323Effort: mediumclaude-transcript-v4 · claude-observed-turn-v192.9 tok/sMin 88.5 tok/s · Max 97.3 tok/sSlow-end p10: 89.4 tok/sa few contributors · a few turns · 10+ responses88.4 tok/sMin 84.2 tok/s · Max 92.6 tok/sSlow-end p10: 85.1 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-fable-5-2Anthropic · Claude Code 2.1.325Effort: highclaude-transcript-v4 · claude-observed-turn-v184.2 tok/sMin 19.8 tok/s · Max 113 tok/sSlow-end p10: 81.9 tok/s10+ contributors · 30+ turns · 1k+ responses35.8 tok/sMin 8.1 tok/s · Max 87.2 tok/sSlow-end p10: 13.2 tok/s10+ contributors · 30+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors30+ turns
claude-fable-5-2Anthropic · Claude Code 2.1.326Effort: highclaude-transcript-v4 · claude-observed-turn-v149 tok/sMin 5 tok/s · Max 86.1 tok/sSlow-end p10: 39.1 tok/sa few contributors · 10+ turns · 100+ responses39.2 tok/sMin 5.6 tok/s · Max 72.2 tok/sSlow-end p10: 25.6 tok/sa few contributors · 30+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributors30+ turns
claude-opus-5-5Anthropic · Claude Code 2.1.317Effort: highclaude-transcript-v4 · claude-observed-turn-v197.4 tok/sMin 67.7 tok/s · Max 137 tok/sSlow-end p10: 91.1 tok/sa few contributors · 30+ turns · 100+ responses72.1 tok/sMin 12.6 tok/s · Max 120 tok/sSlow-end p10: 69.6 tok/sa few contributors · 30+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributors30+ turns
claude-opus-5-5Anthropic · Claude Code 2.1.317Effort: mediumclaude-transcript-v4 · claude-observed-turn-v1117 tok/sMin 92.8 tok/s · Max 130 tok/sSlow-end p10: 113 tok/sa few contributors · a few turns · 10+ responses97.9 tok/sMin 77.6 tok/s · Max 108 tok/sSlow-end p10: 94.9 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-opus-5-5Anthropic · Claude Code 2.1.321Effort: highclaude-transcript-v4 · claude-observed-turn-v193.6 tok/sMin 76.9 tok/s · Max 107 tok/sSlow-end p10: 82.6 tok/sa few contributors · 10+ turns · 100+ responses81.3 tok/sMin 61.8 tok/s · Max 98.5 tok/sSlow-end p10: 67.2 tok/sa few contributors · 10+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributors10+ turns
claude-opus-5-5Anthropic · Claude Code 2.1.323Effort: highclaude-transcript-v4 · claude-observed-turn-v199.2 tok/sMin 85.6 tok/s · Max 122 tok/sSlow-end p10: 94.1 tok/s10+ contributors · 30+ turns · 100+ responses59.3 tok/sMin 17.6 tok/s · Max 97.1 tok/sSlow-end p10: 35.6 tok/s10+ contributors · 30+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors30+ turns
claude-opus-5-5Anthropic · Claude Code 2.1.325Effort: xhighclaude-transcript-v4 · claude-observed-subagent-turn-v192.6 tok/sMin 85.1 tok/s · Max 103 tok/sSlow-end p10: 92.6 tok/sa few contributors · a few turns · 100+ responses80.7 tok/sMin 70.5 tok/s · Max 96.7 tok/sSlow-end p10: 80.7 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-opus-5-5Anthropic · Claude Code 2.1.325Effort: highclaude-transcript-v4 · claude-observed-turn-v185.1 tok/sMin 73 tok/s · Max 95.4 tok/sSlow-end p10: 78.4 tok/s10+ contributors · 10+ turns · 100+ responses35.1 tok/sMin 8.7 tok/s · Max 79.3 tok/sSlow-end p10: 9.3 tok/s10+ contributors · 10+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors10+ turns
claude-opus-5-5Anthropic · Claude Code 2.1.325Effort: lowclaude-transcript-v4 · claude-observed-turn-v181 tok/sMin 71.1 tok/s · Max 92.4 tok/sSlow-end p10: 81 tok/sa few contributors · a few turns · 100+ responses56.4 tok/sMin 12.8 tok/s · Max 80.7 tok/sSlow-end p10: 56.4 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-opus-5-5Anthropic · Claude Code 2.1.325Effort: mediumclaude-transcript-v4 · claude-observed-turn-v185.8 tok/sMin 69.1 tok/s · Max 110 tok/sSlow-end p10: 85.8 tok/sa few contributors · 10+ turns · 100+ responses74.5 tok/sMin 16.1 tok/s · Max 86.4 tok/sSlow-end p10: 74.5 tok/sa few contributors · 10+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributors10+ turns
claude-opus-5-5Anthropic · Claude Code 2.1.325Effort: xhighclaude-transcript-v4 · claude-observed-turn-v195.8 tok/sMin 87.1 tok/s · Max 105 tok/sSlow-end p10: 92.2 tok/sa few contributors · a few turns · 100+ responses67.3 tok/sMin 42.7 tok/s · Max 87.7 tok/sSlow-end p10: 63.9 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-opus-5-5Anthropic · Claude Code 2.1.326Effort: highclaude-transcript-v4 · claude-observed-turn-v169.7 tok/sMin 57.2 tok/s · Max 99.2 tok/sSlow-end p10: 59.7 tok/sa few contributors · a few turns · 100+ responses32.8 tok/sMin 11.9 tok/s · Max 82.5 tok/sSlow-end p10: 16.1 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-sonnet-5-5Anthropic · Claude Code 2.1.317Effort: mediumclaude-transcript-v4 · claude-observed-subagent-turn-v1119 tok/sMin 87.1 tok/s · Max 150 tok/sSlow-end p10: 112 tok/sa few contributors · 10+ turns · 100+ responses55.8 tok/sMin 26.8 tok/s · Max 89.8 tok/sSlow-end p10: 41.1 tok/sa few contributors · 10+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributors10+ turns
claude-sonnet-5-5Anthropic · Claude Code 2.1.321Effort: mediumclaude-transcript-v4 · claude-observed-subagent-turn-v1125 tok/sMin 65.1 tok/s · Max 146 tok/sSlow-end p10: 112 tok/s10+ contributors · 10+ turns · 1k+ responses64.5 tok/sMin 14.9 tok/s · Max 92.5 tok/sSlow-end p10: 57.5 tok/s10+ contributors · 10+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors10+ turns
claude-sonnet-5-5Anthropic · Claude Code 2.1.323Effort: mediumclaude-transcript-v4 · claude-observed-subagent-turn-v1124 tok/sMin 119 tok/s · Max 148 tok/sSlow-end p10: 120 tok/sa few contributors · a few turns · 100+ responses24.3 tok/sMin 19.3 tok/s · Max 29.1 tok/sSlow-end p10: 20.3 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-sonnet-5-5Anthropic · Claude Code 2.1.325Effort: mediumclaude-transcript-v4 · claude-observed-subagent-turn-v195.4 tok/sMin 48.4 tok/s · Max 157 tok/sSlow-end p10: 57.2 tok/sa few contributors · a few turns · 100+ responses37.2 tok/sMin 3.6 tok/s · Max 122 tok/sSlow-end p10: 4.2 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
claude-sonnet-5-5Anthropic · Claude Code 2.1.326Effort: mediumclaude-transcript-v4 · claude-observed-subagent-turn-v1155 tok/sMin 137 tok/s · Max 169 tok/sSlow-end p10: 148 tok/sa few contributors · a few turns · 30+ responses69.9 tok/sMin 17.4 tok/s · Max 75.9 tok/sSlow-end p10: 48.9 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
gemini-3.8-flash-nGoogle · Antigravity unknownEffort: unknownantigravity-conversation-v1 · antigravity-observed-execution-v158 tok/sMin 47.1 tok/s · Max 60.4 tok/sSlow-end p10: 56.1 tok/sa few contributors · a few turns · 10+ responses43.5 tok/sMin 35.3 tok/s · Max 45.2 tok/sSlow-end p10: 42 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
gemini-4-hydrogenGoogle · Antigravity unknownEffort: xhighantigravity-conversation-v1 · antigravity-observed-execution-v155.5 tok/sMin 51.9 tok/s · Max 73 tok/sSlow-end p10: 55.5 tok/sa few contributors · a few turns · 30+ responses43.7 tok/sMin 42.3 tok/s · Max 52.5 tok/sSlow-end p10: 43.7 tok/sa few contributors · a few turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough dataa few contributorsa few turns
gpt-6-astraOpenAI · Codex 0.169.0-alpha.11.1Effort: mediumcodex-rollout-v2 · turn-v125.8 tok/sMin 10.1 tok/s · Max 33.6 tok/sSlow-end p10: 25.8 tok/sa few contributors · 10+ turns · 30+ responses13.6 tok/sMin 7.1 tok/s · Max 24.7 tok/sSlow-end p10: 13.6 tok/sa few contributors · 10+ turns4.6 sMin 2.8 s · Max 38.2 sSlow-end p95: 4.6 sa few contributors · 10+ turnsa few contributors10+ turns
gpt-6-astraUnknown / not reported · Codex 0.167.0-alpha.4.1Effort: maxcodex-rollout-v2 · turn-v126.1 tok/sMin 22.2 tok/s · Max 30 tok/sSlow-end p10: 26.1 tok/sa few contributors · a few turns · 100+ responses25 tok/sMin 20.7 tok/s · Max 29.3 tok/sSlow-end p10: 25 tok/sa few contributors · a few turns3.7 sMin 3.4 s · Max 4 sSlow-end p95: 3.7 sa few contributors · a few turnsa few contributorsa few turns
gpt-6-lunaOpenAI · Codex 0.163.0-alpha.10.2Effort: highcodex-rollout-v2 · turn-v148.9 tok/sMin 40 tok/s · Max 57.5 tok/sSlow-end p10: 45.8 tok/s10+ contributors · 100+ turns · 10k+ responses44.9 tok/sMin 28.7 tok/s · Max 54.4 tok/sSlow-end p10: 42.5 tok/s10+ contributors · 100+ turns5.8 sMin 2.3 s · Max 9.4 sSlow-end p95: 7.4 s10+ contributors · 100+ turns10+ contributors100+ turns
gpt-6.1-solOpenAI · Codex 0.146.0-alpha.9.2Effort: lowcodex-rollout-v2 · turn-v136.3 tok/sMin 27.7 tok/s · Max 47.6 tok/sSlow-end p10: 33.2 tok/s10+ contributors · 100+ turns · 100+ responses16.8 tok/sMin 12.9 tok/s · Max 23.9 tok/sSlow-end p10: 15.9 tok/s10+ contributors · 100+ turns4.6 sMin 2.4 s · Max 8.8 sSlow-end p95: 7.3 s10+ contributors · 100+ turns10+ contributors100+ turns
gpt-6.1-solOpenAI · Codex 0.155.0-alpha.16.4Effort: highcodex-rollout-v2 · turn-v134.1 tok/sMin 32.5 tok/s · Max 35.6 tok/sSlow-end p10: 32.6 tok/sa few contributors · a few turns · 100+ responses26 tok/sMin 14.4 tok/s · Max 32.2 tok/sSlow-end p10: 17.4 tok/sa few contributors · a few turns3.4 sMin 1.7 s · Max 18.2 sSlow-end p95: 15.7 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solOpenAI · Codex 0.163.0-alpha.10.2Effort: highcodex-rollout-v2 · turn-v127.9 tok/sMin 21.7 tok/s · Max 49.3 tok/sSlow-end p10: 23.9 tok/s10+ contributors · 100+ turns · 1k+ responses22.2 tok/sMin 0.2 tok/s · Max 36.3 tok/sSlow-end p10: 15.5 tok/s10+ contributors · 100+ turns12.1 sMin 2.3 s · Max 23.9 sSlow-end p95: 21.2 s10+ contributors · 100+ turns10+ contributors100+ turns
gpt-6.1-solOpenAI · Codex 0.163.0-alpha.10.2Effort: mediumcodex-rollout-v2 · turn-v124.5 tok/sMin 20.4 tok/s · Max 31 tok/sSlow-end p10: 23.1 tok/s10+ contributors · 30+ turns · 1k+ responses16.2 tok/sMin 8.4 tok/s · Max 25.9 tok/sSlow-end p10: 10.4 tok/s10+ contributors · 30+ turns4.2 sMin 2.7 s · Max 6.7 sSlow-end p95: 5.4 s10+ contributors · 30+ turns10+ contributors30+ turns
gpt-6.1-solOpenAI · Codex 0.164.0-alpha.11.1Effort: highcodex-rollout-v2 · turn-v137.6 tok/sMin 37.6 tok/s · Max 37.6 tok/sSlow-end p10: 37.6 tok/sa few contributors · a few turns · 10+ responses22.3 tok/sMin 22.3 tok/s · Max 22.3 tok/sSlow-end p10: 22.3 tok/sa few contributors · a few turns2.7 sMin 2.7 s · Max 2.7 sSlow-end p95: 2.7 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solOpenAI · Codex 0.166.0-alpha.5.1Effort: highcodex-rollout-v2 · turn-v132.8 tok/sMin 26.3 tok/s · Max 43.2 tok/sSlow-end p10: 28.3 tok/s10+ contributors · 100+ turns · 1k+ responses28.1 tok/sMin 18.9 tok/s · Max 35.8 tok/sSlow-end p10: 24.2 tok/s10+ contributors · 100+ turns4.9 sMin 2 s · Max 28.4 sSlow-end p95: 11.5 s10+ contributors · 100+ turns10+ contributors100+ turns
gpt-6.1-solOpenAI · Codex 0.166.0-alpha.5.1Effort: mediumcodex-rollout-v2 · turn-v129.6 tok/sMin 22.6 tok/s · Max 33.8 tok/sSlow-end p10: 26.2 tok/s10+ contributors · 30+ turns · 1k+ responses21.2 tok/sMin 16.2 tok/s · Max 26.7 tok/sSlow-end p10: 17.9 tok/s10+ contributors · 30+ turns5.8 sMin 2.9 s · Max 49 sSlow-end p95: 30.1 s10+ contributors · 30+ turns10+ contributors30+ turns
gpt-6.1-solOpenAI · Codex 0.166.0-alpha.6Effort: highcodex-rollout-v2 · turn-v130.6 tok/sMin 16.6 tok/s · Max 46 tok/sSlow-end p10: 28.2 tok/sa few contributors · 100+ turns · 1k+ responses17.5 tok/sMin 5.3 tok/s · Max 38.4 tok/sSlow-end p10: 15.2 tok/sa few contributors · 100+ turns7.4 sMin 1.7 s · Max 26 sSlow-end p95: 8 sa few contributors · 100+ turnsa few contributors100+ turns
gpt-6.1-solOpenAI · Codex 0.167.0-alpha.3.2Effort: highcodex-rollout-v2 · turn-v132.3 tok/sMin 24.6 tok/s · Max 46.1 tok/sSlow-end p10: 30.5 tok/s10+ contributors · 100+ turns · 1k+ responses24.9 tok/sMin 10.2 tok/s · Max 39 tok/sSlow-end p10: 20.4 tok/s10+ contributors · 100+ turns4.3 sMin 2.4 s · Max 19.8 sSlow-end p95: 5.2 s10+ contributors · 100+ turns10+ contributors100+ turns
gpt-6.1-solOpenAI · Codex 0.167.0-alpha.3.2Effort: maxcodex-rollout-v2 · turn-v111.7 tok/sMin 7.3 tok/s · Max 16.2 tok/sSlow-end p10: 11.7 tok/sa few contributors · a few turns · 30+ responses8.3 tok/sMin 7 tok/s · Max 9.6 tok/sSlow-end p10: 8.3 tok/sa few contributors · a few turns89.2 sMin 17.8 s · Max 160.5 sSlow-end p95: 89.2 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solOpenAI · Codex 0.167.0-alpha.3.2Effort: xhighcodex-rollout-v2 · turn-v116.8 tok/sMin 7.6 tok/s · Max 22.2 tok/sSlow-end p10: 16.8 tok/sa few contributors · a few turns · 100+ responses14.3 tok/sMin 5.2 tok/s · Max 22.2 tok/sSlow-end p10: 14.3 tok/sa few contributors · a few turns22.8 sMin 10.8 s · Max 35.8 sSlow-end p95: 22.8 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solOpenAI · Codex 0.169.0-alpha.11.1Effort: highcodex-rollout-v2 · turn-v129.1 tok/sMin 17.4 tok/s · Max 34.7 tok/sSlow-end p10: 25.4 tok/s10+ contributors · 30+ turns · 1k+ responses23.9 tok/sMin 16.2 tok/s · Max 31.1 tok/sSlow-end p10: 21 tok/s10+ contributors · 30+ turns3.9 sMin 2.2 s · Max 27.7 sSlow-end p95: 13.9 s10+ contributors · 30+ turns10+ contributors30+ turns
gpt-6.1-solOpenAI · Codex 0.169.0-alpha.11.1Effort: xhighcodex-rollout-v2 · turn-v125 tok/sMin 22.7 tok/s · Max 27.5 tok/sSlow-end p10: 23.2 tok/sa few contributors · a few turns · 100+ responses18.9 tok/sMin 8.8 tok/s · Max 26.5 tok/sSlow-end p10: 15.3 tok/sa few contributors · a few turns158.4 sMin 3.4 s · Max 616.6 sSlow-end p95: 297.8 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solUnknown / not reported · Codex 0.167.0-alpha.4.1Effort: highcodex-rollout-v2 · turn-v134.6 tok/sMin 34.6 tok/s · Max 34.6 tok/sSlow-end p10: 34.6 tok/sa few contributors · a few turns · 30+ responses33.3 tok/sMin 33.3 tok/s · Max 33.3 tok/sSlow-end p10: 33.3 tok/sa few contributors · a few turns5.1 sMin 5.1 s · Max 5.1 sSlow-end p95: 5.1 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solUnknown / not reported · Codex 0.167.0-alpha.4.1Effort: maxcodex-rollout-v2 · turn-v135.9 tok/sMin 28.1 tok/s · Max 43.8 tok/sSlow-end p10: 35.9 tok/sa few contributors · a few turns · 100+ responses35.3 tok/sMin 25 tok/s · Max 42.7 tok/sSlow-end p10: 35.3 tok/sa few contributors · a few turns4.8 sMin 2.1 s · Max 5.7 sSlow-end p95: 4.8 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-solUnknown / not reported · Codex 0.167.0-alpha.4.1Effort: mediumcodex-rollout-v2 · turn-v140.9 tok/sMin 18.1 tok/s · Max 55.4 tok/sSlow-end p10: 40.9 tok/sa few contributors · 30+ turns · 100+ responses33.5 tok/sMin 9.7 tok/s · Max 47.9 tok/sSlow-end p10: 33.5 tok/sa few contributors · 30+ turns3.1 sMin 1.7 s · Max 13.3 sSlow-end p95: 3.1 sa few contributors · 30+ turnsa few contributors30+ turns
gpt-6.1-solUnknown / not reported · Codex 0.167.0-alpha.4.1Effort: xhighcodex-rollout-v2 · turn-v131.2 tok/sMin 28.1 tok/s · Max 41 tok/sSlow-end p10: 31.2 tok/sa few contributors · a few turns · 100+ responses29.9 tok/sMin 28.1 tok/s · Max 39.6 tok/sSlow-end p10: 29.9 tok/sa few contributors · a few turns6.7 sMin 2.9 s · Max 33 sSlow-end p95: 6.7 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.163.0-alpha.10.2Effort: highcodex-rollout-v2 · turn-v139.2 tok/sMin 25.3 tok/s · Max 49.6 tok/sSlow-end p10: 38.1 tok/sa few contributors · 100+ turns · 1k+ responses19.2 tok/sMin 4.5 tok/s · Max 33 tok/sSlow-end p10: 17.6 tok/sa few contributors · 1k+ turns2.8 sMin 1.6 s · Max 6.5 sSlow-end p95: 2.9 sa few contributors · 1k+ turnsa few contributors1k+ turns
gpt-6.1-terraOpenAI · Codex 0.163.0-alpha.10.2Effort: mediumcodex-rollout-v2 · turn-v143.2 tok/sMin 39 tok/s · Max 52.1 tok/sSlow-end p10: 41.4 tok/sa few contributors · 30+ turns · 100+ responses20.3 tok/sMin 16.5 tok/s · Max 38 tok/sSlow-end p10: 19.6 tok/sa few contributors · 100+ turns2.3 sMin 1.3 s · Max 3.2 sSlow-end p95: 2.6 sa few contributors · 100+ turnsa few contributors100+ turns
gpt-6.1-terraOpenAI · Codex 0.166.0-alpha.5.1Effort: maxcodex-rollout-v2 · turn-v128.6 tok/sMin 4.3 tok/s · Max 45.5 tok/sSlow-end p10: 28.6 tok/sa few contributors · a few turns · 100+ responses23 tok/sMin 3.4 tok/s · Max 45.3 tok/sSlow-end p10: 23 tok/sa few contributors · a few turns20.2 sMin 5.2 s · Max 38.8 sSlow-end p95: 20.2 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.166.0-alpha.5.1Effort: xhighcodex-rollout-v2 · turn-v120.6 tok/sMin 13.8 tok/s · Max 26.4 tok/sSlow-end p10: 20.6 tok/sa few contributors · a few turns · 100+ responses10.8 tok/sMin 7.4 tok/s · Max 15.4 tok/sSlow-end p10: 10.8 tok/sa few contributors · a few turns17.1 sMin 15.5 s · Max 42.2 sSlow-end p95: 17.1 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.166.0-alpha.6Effort: highcodex-rollout-v2 · turn-v129.6 tok/sMin 26.2 tok/s · Max 33 tok/sSlow-end p10: 29.6 tok/sa few contributors · a few turns · a few responses27.2 tok/sMin 23.8 tok/s · Max 30.6 tok/sSlow-end p10: 27.2 tok/sa few contributors · a few turns8.2 sMin 7.2 s · Max 9.1 sSlow-end p95: 8.2 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.166.0-alpha.6Effort: maxcodex-rollout-v2 · turn-v128 tok/sMin 22.7 tok/s · Max 34.4 tok/sSlow-end p10: 28 tok/sa few contributors · a few turns · 100+ responses14.5 tok/sMin 3.7 tok/s · Max 25 tok/sSlow-end p10: 14.5 tok/sa few contributors · a few turns28.5 sMin 12.7 s · Max 122.1 sSlow-end p95: 28.5 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.166.0-alpha.6Effort: xhighcodex-rollout-v2 · turn-v19.2 tok/sMin 3.3 tok/s · Max 15 tok/sSlow-end p10: 9.2 tok/sa few contributors · a few turns · 30+ responses4.9 tok/sMin 2.3 tok/s · Max 7.5 tok/sSlow-end p10: 4.9 tok/sa few contributors · a few turns148.1 sMin 7.7 s · Max 288.5 sSlow-end p95: 148.1 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.169.0-alpha.11.1Effort: mediumcodex-rollout-v2 · turn-v1Not enough dataMin Not enough data · Max Not enough dataSlow-end p10: Not enough dataNot enough dataMin Not enough data · Max Not enough dataSlow-end p10: Not enough data3.1 sMin 3.1 s · Max 3.1 sSlow-end p95: 3.1 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraOpenAI · Codex 0.169.0-alpha.11.1Effort: xhighcodex-rollout-v2 · turn-v152.4 tok/sMin 52.4 tok/s · Max 52.4 tok/sSlow-end p10: 52.4 tok/sa few contributors · a few turns · a few responses23.6 tok/sMin 23.6 tok/s · Max 23.6 tok/sSlow-end p10: 23.6 tok/sa few contributors · a few turns4.5 sMin 4.5 s · Max 4.5 sSlow-end p95: 4.5 sa few contributors · a few turnsa few contributorsa few turns
gpt-6.1-terraUnknown / not reported · Codex 0.167.0-alpha.4.1Effort: mediumcodex-rollout-v2 · turn-v141.3 tok/sMin 37.4 tok/s · Max 47.8 tok/sSlow-end p10: 41.3 tok/sa few contributors · a few turns · 10+ responses32.1 tok/sMin 11 tok/s · Max 45.1 tok/sSlow-end p10: 32.1 tok/sa few contributors · 10+ turns2.3 sMin 1.4 s · Max 4.5 sSlow-end p95: 2.3 sa few contributors · 10+ turnsa few contributors10+ turns
grok-4.7-buildUnknown / not reported · Grok Build unknownEffort: highgrok-session-v2 · grok-observed-work-turn-v153.1 tok/sMin 40.2 tok/s · Max 68.6 tok/sSlow-end p10: 48.2 tok/s10+ contributors · 30+ turns · 100+ responses36.8 tok/sMin 16.2 tok/s · Max 62.7 tok/sSlow-end p10: 30.1 tok/s10+ contributors · 30+ turnsNot enough dataMin Not enough data · Max Not enough dataSlow-end p95: Not enough data10+ contributors30+ turns
Usage breakdownWhat contributors use

What contributors use

Share of reported turns in this time range—not market share or percentage of people. One contributor can report several models and effort levels, so contributor counts must not be added together.

Coding tools

  • Codex82.9%
    1k+ reported turns · 10+ contributors
  • Claude Code15.3%
    100+ reported turns · 10+ contributors
  • Grok Build1.5%
    30+ reported turns · 10+ contributors
  • Antigravity0.2%
    a few reported turns · a few contributors

Models

  • gpt-6.1-terra32.1%
    1k+ reported turns · 10+ contributors
  • gpt-6.1-sol30.7%
    1k+ reported turns · 10+ contributors
  • gpt-6-luna19.6%
    100+ reported turns · 10+ contributors
  • claude-fable-5-210.1%
    100+ reported turns · 10+ contributors
  • claude-opus-5-55.3%
    100+ reported turns · 10+ contributors
  • grok-4.7-build1.5%
    30+ reported turns · 10+ contributors
  • gpt-6-astra0.6%
    10+ reported turns · a few contributors
  • gemini-3.8-flash-n0.1%
    a few reported turns · a few contributors
  • gemini-4-hydrogen0.1%
    a few reported turns · a few contributors

Effort

  • high81.6%
    1k+ reported turns · 10+ contributors
  • medium13.7%
    100+ reported turns · 10+ contributors
  • low3.4%
    100+ reported turns · 10+ contributors
  • xhigh0.7%
    10+ reported turns · a few contributors
  • max0.5%
    10+ reported turns · a few contributors
  • Unknown / not reported0.1%
    a few reported turns · a few contributors

Providers

  • OpenAI81.6%
    1k+ reported turns · 10+ contributors
  • Anthropic15.3%
    100+ reported turns · 10+ contributors
  • Unknown / not reported2.9%
    100+ reported turns · 10+ contributors
  • Google0.2%
    a few reported turns · a few contributors

Coding tool versions

  • codex 0.163.0-alpha.10.258.5%
    1k+ reported turns · 10+ contributors
  • codex 0.167.0-alpha.3.29.1%
    100+ reported turns · 10+ contributors
  • claude-code 2.1.3216.9%
    100+ reported turns · 10+ contributors
  • codex 0.166.0-alpha.5.15.2%
    100+ reported turns · 10+ contributors
  • claude-code 2.1.3253.4%
    100+ reported turns · 10+ contributors
  • codex 0.146.0-alpha.9.23.2%
    100+ reported turns · 10+ contributors
  • codex 0.166.0-alpha.63.1%
    100+ reported turns · a few contributors
  • claude-code 2.1.3232.5%
    100+ reported turns · 10+ contributors
  • codex 0.169.0-alpha.11.12.3%
    30+ reported turns · 10+ contributors
  • claude-code 2.1.3171.7%
    30+ reported turns · a few contributors
  • grok-build unknown1.5%
    30+ reported turns · 10+ contributors
  • codex 0.167.0-alpha.4.11.3%
    30+ reported turns · a few contributors
  • Other published categories1.2%
    30+ reported turns · 10+ contributors

Effort is the setting explicitly recorded by the coding tool, not reasoning-token volume or a measure of quality. Collection begins with Tokrate 0.1.6; older or ambiguous observations stay Unknown. Fast mode is not recorded. Coding tools and inference providers are separate: an unknown provider means the source did not establish the routing provider.

Only categories with at least 1 contributor and 1 reported turn appear. Percentages cover displayed categories separately for each breakdown. This section follows the time range above and includes all reported coding tools, providers and models, regardless of the performance filters.

Last 7 days

As of 2026-11-17, over the last 7 days the fastest coding model on Tokrate is gemini-flash-worker-agy at 145 tok/s (Antigravity, google; a few runs / a few contributors), by observed median response speed.

Window: 2026-11-10 18:55–2026-11-17 18:55 UTC (end exclusive). Top 10 of 20 qualifying model/provider rows.

Ranked by median response speed, in tokens per second. Runs are eligible completed turns; coverage uses approximate bands (a few = 1–9). Tools and your time are excluded. Different workloads, effort settings and main/subagent sessions affect the results; answer quality and visible-text streaming speed are not measured.
RankModelResponse speedCoverageLast observed (UTC)
1GoogleGgemini-flash-worker-agyAntigravityGoogle145tok/sa few runsa few contributors
2GoogleGgemini-4-argon-trialAntigravityGoogle143tok/s10+ runsa few contributors
3AnthropicAclaude-sonnet-5-5Claude CodeAnthropic119tok/s100+ runs30+ contributors
4AnthropicAclaude-sonnet-5-5-highAntigravityProvider not reported109tok/sa few runsa few contributors
5AnthropicAclaude-opus-5-5Claude CodeAnthropic91.1tok/s1k+ runs30+ contributors
6AnthropicAclaude-fable-5-2Claude CodeAnthropic90.3tok/s1k+ runs30+ contributors
7AnthropicAclaude-opus-5-5-highAntigravityProvider not reported89.6tok/sa few runsa few contributors
8GoogleGgemini-4-hydrogenAntigravityGoogle88.3tok/s10+ runsa few contributors
9AnthropicAclaude-sonnet-5-5-mediumAntigravityProvider not reported74.4tok/s10+ runsa few contributors
10GoogleGgemini-4-argon-trialAntigravityProvider not reported65.8tok/sa few runsa few contributors

Explore the 7-day board · Methodology and sample rules

Ask ChatGPT about the community data.

Connect Tokrate’s read-only MCP service to explore published comparisons. No desktop companion needed.

Use Tokrate in ChatGPT

One agent turn, two speeds

A coding turn alternates between the model responding and tools running. Response speed times only the responding.

  1. Your promptnot counted
  2. Responsecounted
  3. Tool runexcluded
  4. Responsecounted
  5. Tool runexcluded
  6. Final answercounted
Response speed
= output tokens ÷ time the model is responding. The first-token wait is included; tools and your time are excluded. Responses under 200 output tokens are not counted.
Turn speed
= output tokens ÷ the whole turn, from your prompt to the final answer. Tool runs and waiting are included.
Timeline: your prompt, a counted response, an excluded tool run, a counted response, an excluded tool run, then the counted final answer. Response speed uses the counted parts; turn speed uses the whole turn.
  • Response speedHeadline

    How fast the model answers. Typical coding responses: tens to a few hundred tok/s; bigger models are usually slower.

  • Turn speedSecondary

    How long your agent’s work takes. It reflects the workload: tool runs and waiting count too, so numbers are far lower than response speed.

  • First tokenCodex source

    The wait before the first output appears, in seconds. Reported by Codex, including supported T3 Code observations parsed from Codex. Claude Code, Grok Build, Antigravity, OpenCode and Kimi Code observations do not report it.

  • Efficiency indicatorIndicator

    Fewer output tokens per request scores higher. 100 = a typical request. Not a speed: it compares the output tokens a model spends per request, including subagent work. Needs Tokrate 0.1.16 or newer.

Your own pace, right on your desktop.

A quiet meter for your real coding sessions. See each model’s response speed, how it compares with the previous day, and your own local history, then compare the community’s observations. Available for Mac, with Windows and Linux alpha builds.

  • Claude CodeAnthropic
  • CodexOpenAI
  • Grok BuildxAI
  • AntigravityNewGoogle
  • OpenCodeNewAny provider
  • Kimi CodeNewMoonshot AI
  • T3 CodeNewT3

Product names are trademarks of their owners; Tokrate is independent and not affiliated with them.

Missing your coding tool? Suggest it:@mvilola on XCalltrin

Only numbers leave your computer.

Tokrate reads your local session logs to time each answer. The text stays on your machine; only the numbers are sent.

No ads or behavioral analytics. Sharing sends performance measurements under a random installation key.

Don’t take our word for it. Ask your AI agent to check the open-source code.

Each model, its own baseline.

Compare each model’s response speed, with every coding tool, provider and recorded effort setting kept separate. Turn speed and first-token waits sit alongside, and you can spot changes in your own sessions.

Real work. Honest numbers.

Different tasks take different time. Tools, context, effort and speed tier affect observations. Tokrate does not measure answer quality or streaming speed.

Questions before you install

Does Tokrate see my conversations?

Tokrate reads your local session logs on your computer to time each answer, but your conversations never leave it. No prompts, responses, code, file paths, session IDs, names or account details are ever uploaded. Sharing is optional: from version 0.1.11 it stays off until you choose “Yes, let's contribute”, and with “Only for local use” no measurements are sent at all (software update checks are a separate switch).

If you do share, each finished turn sends only performance numbers: coding tool, model, inference provider, recorded reasoning effort, client versions, token counts (including the output tokens of subagent work a turn started), durations, a five-minute time bucket and a random sample ID. The server stores your continent, never your country or IP address. Contribution data is not sold and is not used for advertising.

Does Tokrate use my tokens or cost me anything?

No. Tokrate never calls a model and never sends prompts on your behalf. It reads the session logs that Codex, Claude Code, Grok Build, Antigravity, OpenCode and Kimi Code already write on your computer; from Tokrate 0.1.25 it also recognizes supported T3 Code observations from Codex, Claude Code and Antigravity. It calculates speed from the token counts and timestamps in those records, so it adds nothing to your token usage, API bill or subscription limits.

It does not ask for API keys and needs no Tokrate account. The companions and the public board are free, with no ads.

Is Tokrate safe to install?

The Mac app is signed with an Apple Developer ID and notarized by Apple. Windows and Linux builds are early alpha releases: the Windows installer is not Authenticode-signed yet, so Windows may show an unknown-publisher warning. Every release publishes SHA-256 checksums, and in-app updates are signature-verified before you choose to install them.

The companions are open source under the MIT license and contain no advertising or behavioral analytics SDKs. You can read exactly what leaves your computer, or ask your AI agent to review the code for you. AI models have already reviewed it against our privacy promises, and the latest AI audit result is published unedited.

Is Tokrate free?

Yes. The public board at tokrate.dev, the companions for Mac, Windows and Linux, and the public API are free to use, with no account, no ads and no paid tier.

All questions