Time per task: summarize a 30-page report

12,000 tokens of document in, a 600-token summary out. Time = time to first answer + output tokens ÷ output speed.

As of 25 Oct 2026, BenchLeader ranks Celeris-1 first on the Time per task: summarize a 30-page report board at 0.9 s, ahead of Gemini 2.5 Flash-Lite (2.4 s) and DeepSeek V4.1 Flash (2.9 s).

520 models · data as of 25 Oct 2026

1. 1Celeris-10.9 s
2. 2Gemini 2.5 Flash-Lite (none)2.4 s
3. 3DeepSeek V4.1 Flash (none)2.9 s
4. 4Nemotron 3 Nano 30B A3B (none)3.0 s
5. 5Nova Micro3.0 s
6. 6Gemini 2.5 Flash (none)3.3 s
7. 7GPT-4.1 nano3.4 s
8. 8Qwen3 Coder 30B A3B Instruct3.5 s
9. 9Ministral 3 3B3.6 s
10. 10Gemini 3 Flash (none)3.8 s
11. 11Gemini 3.5 Flash (minimal)3.9 s
12. 12GPT-5.4 Mini (none)4.1 s
13. 13Nemotron Nano 12B v2 VL (none)4.1 s
14. 14GPT-5.4 Nano (none)4.2 s
15. 15Qwen3 30B A3B 25074.2 s

SearchAll providersAlibabaAmazonAnt GroupAnthropicArcee AIByteDanceCelerisCohereDeepSeekGoogleIBMInceptionMBZUAIMeituanMetaMicrosoftMiniMaxMistral AIMoonshot AIMultiverse ComputingNous ResearchNVIDIAOpenAIReflectionRekaSapiens AISarvamSpaceXAIStepFunSwiss AIThinking MachinesUpstageXiaomiZhipu AIOpen weightsNew modelsBest per model (244)244 of 244

#	Model	seconds ↓	Index ↕
1	Ling Flash 2.0Ant Groupopen	2.4 min	44.4
2	DeepSeek V4 Flash 0420maxDeepSeekopen	2.2 min	57.0
3	Qwen3.6 PlusAlibaba	1.9 min	58.8
4	GPT-5.4 ProOpenAI	1.5 min	63.3
5	DeepSeek R1 0528DeepSeekopen ↗	1.3 min	49.5
6	Seed-OSS-36B-InstructByteDanceopen ↗	1.2 min	46.8
7	GPT-5.3-CodexxhighOpenAI	1.1 min	63.8
8	Qwen3.8 2.4T A95BAlibabaopen ↗	1.1 min	61.0
9	MiniMax M2.7MiniMaxopen ↗	57.8 s	55.6
10	Qwen3.5-4Bno reasoningAlibabaopen ↗	55.0 s	42.2
11	Qwen3.8-Flash-NextAlibabaopen ↗	52.6 s	60.3
12	MiMo-V2.6-ProXiaomiopen ↗	50.2 s	64.0
13	Claude Opus 4.1thinkingAnthropic	50.1 s	56.4
14	Kimi K2.7 CodeMoonshot AIopen ↗	50.0 s	55.8
15	MiMo-V2.6-FlashXiaomiopen ↗	47.9 s	59.7
16	Ling 3.0 TinyAnt Groupopen	47.9 s	48.7
17	Qwen3.7 PlusAlibaba	46.2 s	58.5
18	North Mini CodeCohereopen	43.6 s	45.9
19	Grok Build 0.1SpaceXAI	41.5 s	54.8
20	MiniMax M2.5MiniMaxopen ↗	41.4 s	53.6
21	GLM 5.3 FlashZhipu AIopen ↗	41.1 s	62.7
22	MiniMax-M2.1MiniMaxopen ↗	40.1 s	53.8
23	Grok 3 minithinkingSpaceXAI	39.7 s	49.4
24	NewKimi K3.1 ProMoonshot AI	38.4 s	66.9
25	DeepSeek V3.2DeepSeekopen ↗	37.4 s	56.6
26	LongCat 2.0Meituanopen ↗	36.6 s	49.4
27	GLM-4.5-AirZhipu AIopen ↗	36.4 s	46.6
28	MiMo-V2.5Xiaomiopen ↗	35.9 s	54.3
29	Granite 4.2 30BIBMopen ↗	35.6 s	48.7
30	Magistral Small 1.2Mistral AIopen	34.7 s	40.3
31	GLM 5 TurboZhipu AI	33.7 s	55.5
32	GPT-5.5 ProxhighOpenAI	33.6 s	65.3
33	GLM 5.3lowZhipu AIopen ↗	32.3 s	55.6
34	Reka Flash 3Rekaopen ↗	31.0 s	38.9
35	GLM-4.5Zhipu AIopen ↗	30.8 s	51.7
36	Granite 4.2 8BIBMopen ↗	30.7 s	46.1
37	Mistral Large 4Mistral AIopen	30.6 s	56.7
38	MiniMax M3MiniMaxopen ↗	30.6 s	56.4
39	o4-minihighOpenAI	29.3 s	50.1
40	MiniMax-M1MiniMaxopen	29.2 s	47.5
41	NewMiniMax M3.1 FlashMiniMaxopen ↗	28.5 s	61.9
42	Granite 3.3 8Bno reasoningIBMopen ↗	28.0 s	39.2
43	MiniMax-M2MiniMaxopen ↗	27.6 s	50.8
44	Llama 3.2 Instruct 11B (Vision)Metaopen ↗	27.5 s	36.5
45	Gemini 2.5 ProGoogle	27.4 s	53.9
46	Granite 4.2 3BIBMopen ↗	26.9 s	44.8
47	Qwen2.5 72BAlibabaopen ↗	25.8 s	42.2
48	Gemini 3.1 ProGoogle	25.6 s	60.5
49	Ring-2.6-1TAnt Groupopen	25.3 s	49.3
50	Mixtral 8x7BMistral AIopen	25.1 s	31.7
51	Qwen3 VL 30B A3B InstructthinkingAlibabaopen ↗	24.8 s	44.0
52	Agnes 2.5 Pro BetaSapiens AI	24.4 s	59.1
53	DeepSeek V3.1thinkingDeepSeekopen ↗	24.4 s	48.1
54	Gemma 3 27BGoogleopen ↗	23.9 s	38.9
55	Kimi K2 ThinkingthinkingMoonshot AIopen ↗	23.4 s	52.4
56	GLM-4.6no reasoningZhipu AIopen ↗	22.4 s	46.7
57	Qwen2.5 Coder 32B InstructAlibabaopen ↗	21.6 s	42.1
58	DeepSeek R1DeepSeekopen ↗	21.5 s	46.8
59	Step 3.5 FlashStepFunopen ↗	21.3 s	53.7
60	GPT-4 TurboOpenAI	21.2 s	40.9

Show 100 more

Cite as: BenchLeader, “Time per task: summarize a 30-page report”, data as of 25 Oct 2026, https://www.benchleader.com/leaderboards/time-summarize-report.