Real-world LLM speed leaderboard

Anonymous real-world speed results from TOKRACE, annotated by sample confidence. Current 9 models · 516 runs

ModelMedian tok/sTTFTPeak
🥇kimi-for-coding
api.kimi.com· 119 runsStable sample
188
avg 235
1.48s
900
🥈step-3.7-flash
api.stepfun.com· 69 runsStable sample
170
avg 178
1.99s
2,067
🥉deepseek-v4-flash
api.deepseek.com· 76 runsStable sample
142
avg 133
0.79s
346
4mimo-v2.5
api.xiaomimimo.com· 39 runsUsable signal
86
avg 89
4.71s
514
5deepseek-v4-pro
api.deepseek.com· 48 runsUsable signal
80
avg 85
1.39s
293
6MiniMax-M3
api.minimaxi.com· 3 runsLow sample
80
avg 86
1.40s
179
7glm-5.1
open.bigmodel.cn· 51 runsStable sample
53
avg 61
4.70s
594
8glm-5.2
open.bigmodel.cn· 76 runsStable sample
48
avg 58
3.98s
475
9mimo-v2.5-pro
api.xiaomimimo.com· 35 runsUsable signal
40
avg 43
2.03s
182

Speed vs cost

Full pricing table →

Median output speed against output price. Top-left is best (fast & cheap); the dashed line marks the Pareto frontier — models not beaten on both speed and price.

052103155207$0.30$1$3Output $/1M (log, ← cheaper)Median tok/s (faster ↑)↖ fast & cheap = bestkimi-for-coding · api.kimi.com speed 188 tok/s · output $4/1M · Pareto frontierkimi-for-codingstep-3.7-flash · api.stepfun.com speed 170 tok/s · output $1/1M · Pareto frontierstep-3.7-flashdeepseek-v4-flash · api.deepseek.com speed 142 tok/s · output $0.28/1M · Pareto frontierdeepseek-v4-flashmimo-v2.5 · api.xiaomimimo.com speed 86 tok/s · output $0.28/1M · Pareto frontiermimo-v2.5deepseek-v4-pro · api.deepseek.com speed 80 tok/s · output $0.87/1MMiniMax-M3 · api.minimaxi.com speed 80 tok/s · output $1/1Mglm-5.1 · open.bigmodel.cn speed 53 tok/s · output $4/1Mglm-5.2 · open.bigmodel.cn speed 48 tok/s · output $4/1Mmimo-v2.5-pro · api.xiaomimimo.com speed 40 tok/s · output $0.28/1M

· Cost uses each model's output price per 1M tokens (CNY converted to USD); speed is the median from the leaderboard above.

Popular comparisons

Head-to-head speed pages for the top models on the board.

kimi-for-coding vs step-3.7-flashdeepseek-v4-flash vs kimi-for-codingkimi-for-coding vs mimo-v2.5deepseek-v4-pro vs kimi-for-codingkimi-for-coding vs MiniMax-M3glm-5.1 vs kimi-for-codingglm-5.2 vs kimi-for-codingdeepseek-v4-flash vs step-3.7-flashmimo-v2.5 vs step-3.7-flashdeepseek-v4-pro vs step-3.7-flashMiniMax-M3 vs step-3.7-flashglm-5.1 vs step-3.7-flashglm-5.2 vs step-3.7-flashdeepseek-v4-flash vs mimo-v2.5deepseek-v4-flash vs deepseek-v4-prodeepseek-v4-flash vs MiniMax-M3deepseek-v4-flash vs glm-5.1deepseek-v4-flash vs glm-5.2deepseek-v4-pro vs mimo-v2.5mimo-v2.5 vs MiniMax-M3glm-5.1 vs mimo-v2.5glm-5.2 vs mimo-v2.5deepseek-v4-pro vs MiniMax-M3deepseek-v4-pro vs glm-5.1deepseek-v4-pro vs glm-5.2glm-5.1 vs MiniMax-M3glm-5.2 vs MiniMax-M3glm-5.1 vs glm-5.2

· Data comes from voluntary anonymous sharing and contains speed metrics only, not prompts or API keys · Updates every 5 minutes

· Speed is affected by network, time of day and provider load; different endpoints for the same model are tracked separately