Production Benchmarks

Real performance data from production traffic. Not synthetic benchmarks.

Live data2.4B tokens across 87 models from 2,667 developers

#	Model	Tokens	Speed	Uptime
1	qwen-3.6-plusby Alibaba	524.3M	54tok/s	100%
2	gemma-4-31bby Google	391.3M	23tok/s	100%
3	deepseek-v4-flashby DeepSeek	348.5M	37tok/s	100%
4	deepseek-v3by DeepSeek	180.2M	14tok/s	100%
5	gpt-oss-120bby OpenAI	123.2M	29tok/s	100%
6	minimax-m2.7by MiniMax	118.5M	51tok/s	100%
7	gemini-2.5-flashby Google	90.8M	57tok/s	99%
8	gemini-3-flashby Google	86.9M	49tok/s	99%
9	qwen-3-coderby Alibaba	78.9M	58tok/s	99%
10	llama-3.3-70bby Meta	70.9M	190tok/s	99%
11	deepseek-v4-proby DeepSeek	58.8M	37tok/s	98%
12	glm-4.7-flashby Zhipu AI	50.8M	49tok/s	100%
13	minimax-m2.5by MiniMax	48.1M	46tok/s	100%
14	qwen-3-32bby Alibaba	37.5M	41tok/s	99%
15	kimi-k2.5by Moonshot	34.8M	27tok/s	100%
16	deepseek-r1by DeepSeek	20.1M	21tok/s	100%
17	gemini-3.5-flashby Google	15.2M	128tok/s	85%
18	glm-4.5-airby Zhipu AI	14.8M	44tok/s	100%
19	glm-5.1by Zhipu AI	14.7M	63tok/s	100%
20	glm-5.2by Zhipu AI	8.6M	75tok/s	100%
21	kimi-k2.6by Moonshot	8.4M	28tok/s	100%
22	gemini-3-flash-audioby Google	3.8M	—	100%
23	qwen-3.7-maxby Alibaba	3.8M	50tok/s	100%
24	grok-4.3by xAI	3.6M	112tok/s	100%
25	kimi-k2.7-codeby Moonshot	2.7M	34tok/s	100%
26	qwen-3.7-plusby Alibaba	2.1M	52tok/s	100%
27	qwen3.7-flashby Alibaba	1.4M	132tok/s	100%
28	grok-buildby xAI	909.2K	164tok/s	100%
29	whisper-v3-turboby OpenAI	677.3K	—	100%
30	A claude-sonnet-4-6by Anthropic	620.4K	28tok/s	100%
31	gpt-5.6-terraby OpenAI	527.5K	51tok/s	100%
32	nemotron-3-ultra-550bby NVIDIA	464.1K	52tok/s	100%
33	gpt-5.6-lunaby OpenAI	454.1K	103tok/s	100%
34	minimax-m3by MiniMax	379.5K	29tok/s	100%
35	sonarby Perplexity	343.6K	32tok/s	100%
36	kling-3-proby Kuaishou	330.8K	—	100%
37	gemini-3.6-flashby Google	302.6K	107tok/s	100%
38	seedance-2-proby ByteDance	253.6K	—	100%
39	grok-4.20-multi-agentby xAI	187.3K	453tok/s	100%
40	kling-3-pro-audioby Kuaishou	160.8K	—	100%
41	kling-2.5-proby Kuaishou	149.4K	—	100%
42	seedance-2-fastby ByteDance	138.7K	—	100%
43	flux-1.1-ultraby Black Forest Labs	112.4K	—	100%
44	ideogram-v3by Ideogram	111.8K	—	100%
45	flux-kontext-proby Black Forest Labs	104.1K	—	100%
46	kimi-k3by Moonshot	99.7K	23tok/s	100%
47	A claude-opus-4-7by Anthropic	89.4K	26tok/s	100%
48	gpt-5.6-luna-proby OpenAI	40.2K	88tok/s	100%
49	gpt-5.6-terra-proby OpenAI	39.9K	95tok/s	100%
50	gpt-5.6-sol-proby OpenAI	39.3K	64tok/s	100%
51	A claude-haiku-4-5by Anthropic	31.4K	44tok/s	100%
52	step-3.7-flashby StepFun	20.3K	58tok/s	100%
53	gpt-4o-mini-transcribe-2025-12-15by OpenAI	18.9K	—	100%
54	llama-4-maverickby Meta	18.7K	33tok/s	100%
55	sonar-proby Perplexity	16.0K	18tok/s	97%
56	muse-spark-1.1by Meta	9.4K	169tok/s	100%
57	grok-4.5by xAI	9.2K	59tok/s	100%
58	gemini-3.5-flash-liteby Google	8.0K	72tok/s	100%
59	grok-4.20by xAI	5.4K	64tok/s	100%
60	A claude-opus-5by Anthropic	4.5K	20tok/s	100%
61	A claude-sonnet-5by Anthropic	3.2K	26tok/s	100%
62	recraft-v3by Recraft	2.9K	—	100%
63	A claude-opus-5-fastby Anthropic	2.9K	37tok/s	100%
64	gpt-5.6-solby OpenAI	1.9K	37tok/s	100%
65	embeddinggemma-300mby Google	1.1K	—	100%
66	eleven-flash-v2-5by ElevenLabs	112	—	100%
67	hailuo-02-512pby MiniMax	56	—	100%
68	minimax-voice-designby MiniMax	56	—	100%
69	hailuo-02-1080pby MiniMax	56	—	100%
70	flux-2-proby Black Forest Labs	56	—	100%
71	minimax-voice-cloneby MiniMax	56	—	100%
72	minimax-musicby MiniMax	56	—	100%
73	elevenlabs-sfxby ElevenLabs	56	—	100%
74	elevenlabs-musicby ElevenLabs	56	—	100%
75	minimax-music-proby MiniMax	56	—	100%
76	minimax-speech-hdby MiniMax	56	—	100%
77	eleven-turbo-v2-5by ElevenLabs	56	—	100%
78	recraft-v4-vector-proby Recraft	56	—	100%
79	eleven-multilingual-v2by ElevenLabs	56	—	100%
80	recraft-v4-vectorby Recraft	56	—	100%
81	recraft-v4-proby Recraft	56	—	100%
82	minimax-speech-turboby MiniMax	56	—	100%
83	minimax-image-01by MiniMax	56	—	100%
84	hailuo-02-768pby MiniMax	56	—	100%
85	recraft-v4by Recraft	56	—	100%
86	qwen3-embedding-8bby Alibaba	25	—	100%
87	eleven-v3by ElevenLabs	0	—	100%

Measured from real API traffic. Updated every 5 minutes. Read docs →