Response time: seconds to a complete answer

Median seconds from request to the last token, reasoning included (Artificial Analysis). Time to first answer tells you when output starts; this tells you when it ends.

As of 21 Sept 2026, BenchLeader ranks Celeris-1 first on the Response time: seconds to a complete answer board at 0.9 s, ahead of Gemini 2.5 Flash-Lite (1.9 s) and Gemini 3.7 Flash (2.3 s).

354 models · data as of 21 Sept 2026

Model sizeAll 128Tiny 2Small 14Medium 7Large 30

• OpenAI31
• Anthropic18
• Alibaba15
• Google12
• Mistral AI8
• SpaceXAI7
• DeepSeek6
• Meta5
• Zhipu AI5
• Moonshot AI4
• NVIDIA4
• IBM2
• MiniMax2
• Xiaomi2
• Ant Group1
• Arcee AI1
• CCeleris1
• Meituan1

The dashed staircase is the efficiency frontier: at each price, the best index anyone reaches. A model above and left of the rest gives you more quality per dollar of real work. Size bands and run costs come from Artificial Analysis, which is the only source that publishes what a fixed batch of work actually cost.

1. 1Celeris-10.9 s
2. 2Gemini 2.5 Flash-Lite (none)1.9 s
3. 3Gemini 3.7 Flash (low)2.3 s
4. 4Gemini 2.5 Flash (none)2.8 s
5. 5Nova Micro2.8 s
6. 6Ministral 3 3B2.9 s
7. 7GPT-5.4 Mini (none)3.3 s
8. 8Gemini 3 Flash (none)3.3 s
9. 9Gemini 3.5 Flash (minimal)3.5 s
10. 10GPT-4.1 nano3.6 s
11. 11GPT-4.13.6 s
12. 12GPT-5 nano (minimal)3.7 s
13. 13Nemotron 3 Nano 30B A3B (none)3.8 s
14. 14Nova Lite3.8 s
15. 15GPT-5.4 Nano (none)3.8 s

SearchAll providersAlibabaAmazonAnt GroupAnthropicArcee AIByteDanceCelerisCohereDeepSeekGoogleIBMInceptionMetaMicrosoftMiniMaxMistral AIMoonshot AIMultiverse ComputingNous ResearchNVIDIAOpenAISapiens AISpaceXAIStepFunThinking MachinesUnknownUpstageXiaomiZhipu AIOpen weightsNew modelsBest per model (191)191 of 191

#	Model	seconds ↓	Thinking ↕	Index ↕
1	Claude Fable 5thinkingAnthropic	2.4 min	–	70.8
2	Qwen3.6 PlusAlibaba	1.8 min	1.6 min	58.5
3	MiMo-V2.5Xiaomiopen ↗	1.3 min	59.1 s	55.8
4	Ling Flash 2.0Ant Groupopen	1.2 min	–	45.0
5	Gemini 3.1 ProGoogle	1.1 min	–	63.8
6	Qwen3 8maxAlibaba	1.1 min	51.3 s	66.5
7	Qwen3.8 2.4T A95BAlibabaopen ↗	1.1 min	50.0 s	65.1
8	ByteDance-Seed/Seed-OSS-36B-InstructByteDanceopen	1.1 min	49.8 s	47.4
9	GPT-5.3-CodexxhighOpenAI	1.1 min	–	65.3
10	Kimi K3maxMoonshot AIopen ↗	1.0 min	46.6 s	67.1
11	Ling 3.0 TinyAnt Groupopen	51.7 s	38.8 s	50.0
12	Granite 3.3 8Bno reasoningIBMopen ↗	51.4 s	–	39.4
13	Grok Build 0.1SpaceXAI	49.0 s	38.8 s	54.9
14	Kimi K2.7 CodeMoonshot AIopen ↗	46.7 s	35.7 s	55.8
15	Granite 4.0 H SmallIBMopen ↗	45.5 s	–	40.6
16	GLM 5.3maxZhipu AIopen ↗	44.0 s	32.8 s	65.8
17	Qwen3maxAlibaba	43.8 s	31.8 s	53.9
18	Qwen3.8-Flash-NextAlibabaopen ↗	43.7 s	32.8 s	61.5
19	Claude Opus 4.8maxAnthropic	43.0 s	–	64.1
20	Solar Pro 4Upstage	40.8 s	30.9 s	55.4
21	Qwen3 6maxAlibaba	39.3 s	28.8 s	63.3
22	o4-minihighOpenAI	39.3 s	–	50.7
23	Qwen3.5-4Bno reasoningAlibabaopen ↗	39.3 s	–	42.8
24	Qwen3.7 PlusAlibaba	38.5 s	29.0 s	59.8
25	MiniMax M2.7MiniMaxopen ↗	38.2 s	30.5 s	56.0
26	Granite 4.2 30BIBMopen ↗	35.0 s	27.3 s	49.8
27	GLM-4.5-AirZhipu AIopen ↗	33.5 s	24.7 s	47.1
28	Grok 3 minithinkingSpaceXAI	32.0 s	22.9 s	50.2
29	Muse Spark 1.3xhighMeta	30.9 s	7.1 s	67.9
30	Muse Spark 1.2xhighMeta	30.8 s	9.8 s	64.0
31	NewStep 5 PreviewStepFun	30.2 s	21.7 s	64.9
32	Phi 4 MultimodalMicrosoftopen	30.1 s	–	–
33	GLM 5.3 FlashZhipu AIopen ↗	29.1 s	21.4 s	63.9
34	Granite 4.2 8BIBMopen ↗	29.0 s	22.5 s	46.8
35	Muse GlimmerhighMetaopen ↗	28.4 s	21.9 s	52.8
36	Gemini 3.6 FlashhighGoogle	28.1 s	–	61.7
37	Gemini 2.5 ProGoogle	27.5 s	–	54.4
38	MiniMax M2.5MiniMaxopen ↗	27.3 s	20.6 s	54.2
39	Step 3.5 FlashStepFunopen ↗	27.3 s	19.0 s	54.6
40	North Mini CodeCohereopen	27.1 s	21.4 s	46.5
41	Llama 3.2 Instruct 11B (Vision)Metaopen ↗	25.5 s	–	37.1
42	MiniMax-M2MiniMaxopen ↗	25.1 s	18.7 s	55.0
43	Gemini 3.8 FlashhighGoogle	24.1 s	–	64.5
44	Ring-2.6-1TAnt Groupopen	22.4 s	14.9 s	50.0
45	Kimi K2 ThinkingthinkingMoonshot AIopen ↗	21.8 s	16.3 s	52.1
46	Ling 3.0 Flash VLAnt Groupopen ↗	19.9 s	13.6 s	56.6
47	Solar Pro 3Upstage	19.8 s	13.9 s	43.9
48	Phi-4Microsoftopen ↗	19.8 s	–	39.5
49	Mistral Medium 3.5Mistral AIopen	19.7 s	13.9 s	51.2
50	GPT-4 TurboOpenAI	19.4 s	–	41.5
51	Llama Nemotron Super 49B v1.5thinkingNVIDIAopen	19.1 s	12.4 s	44.0
52	ULing 3.0 Flash FinUnknownopen	18.4 s	12.6 s	55.3
53	GLM 4.5Vno reasoningZhipu AIopen ↗	17.7 s	–	39.0
54	Quasar 438BMultiverse Computing	17.4 s	12.9 s	57.1
55	Claude Fable 5.1mediumAnthropic	17.2 s	–	69.5
56	GLM-4.6no reasoningZhipu AIopen ↗	17.2 s	–	47.3
57	Agnes 2.5 Pro AlphaSapiens AIopen	17.2 s	10.6 s	54.0
58	GPT-5.5 InstantOpenAI	17.2 s	12.9 s	59.0
59	Qwen3 7maxAlibaba	16.8 s	12.0 s	60.9
60	Nemotron 3 Ultra 550B A55BthinkingNVIDIAopen ↗	16.7 s	11.8 s	57.5

Show 100 more

Cite as: BenchLeader, “Response time: seconds to a complete answer”, https://www.benchleader.com/leaderboards/response-time, data as of 21 Sept 2026.