Time per task: code edit

6,000 tokens of files and instructions in, a 1,500-token diff 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: code edit board at 1.5 s, ahead of Gemini 2.5 Flash-Lite (5.4 s) and K2 Horizon 375B A23B (5.9 s).

520 models · data as of 25 Oct 2026

1. 1Celeris-11.5 s
2. 2Gemini 2.5 Flash-Lite (none)5.4 s
3. 3K2 Horizon 375B A23B5.9 s
4. 4Mercury 2.55.9 s
5. 5DeepSeek V4.1 Flash (none)6.0 s
6. 6Nova Micro6.2 s
7. 7Nemotron 3 Nano 30B A3B (none)6.4 s
8. 8Qwen3 Coder 30B A3B Instruct7.2 s
9. 9GPT-4.1 nano7.4 s
10. 10Gemini 2.5 Flash (none)7.7 s
11. 11Ministral 3 3B7.9 s
12. 12Gemini 3 Flash (none)8.0 s
13. 13Qwen3.5 Omni Flash8.2 s
14. 14Gemini 3.1 Flash Lite8.4 s
15. 15Gemini 3.5 Flash (minimal)8.5 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	6.0 min	44.4
2	GPT-5.4 ProOpenAI	3.7 min	63.3
3	DeepSeek V4 Flash 0420maxDeepSeekopen	2.5 min	57.0
4	Qwen3.5-4Bno reasoningAlibabaopen ↗	2.3 min	42.2
5	Qwen3.6 PlusAlibaba	2.1 min	58.8
6	Claude Opus 4.1thinkingAnthropic	2.0 min	56.4
7	DeepSeek R1 0528DeepSeekopen ↗	1.8 min	49.5
8	Seed-OSS-36B-InstructByteDanceopen ↗	1.6 min	46.8
9	DeepSeek V3.2DeepSeekopen ↗	1.5 min	56.6
10	Qwen3.8 2.4T A95BAlibabaopen ↗	1.4 min	61.0
11	GPT-5.3-CodexxhighOpenAI	1.3 min	63.8
12	MiniMax M2.7MiniMaxopen ↗	1.2 min	55.6
13	GLM 5 TurboZhipu AI	1.2 min	55.5
14	Qwen3.8-Flash-NextAlibabaopen ↗	1.2 min	60.3
15	GPT-5.5 ProxhighOpenAI	1.1 min	65.3
16	Llama 3.2 Instruct 11B (Vision)Metaopen ↗	1.1 min	36.5
17	MiMo-V2.6-ProXiaomiopen ↗	1.1 min	64.0
18	Kimi K2.7 CodeMoonshot AIopen ↗	1.1 min	55.8
19	GLM-4.5Zhipu AIopen ↗	1.1 min	51.7
20	Ling 3.0 TinyAnt Groupopen	1.1 min	48.7
21	MiMo-V2.6-FlashXiaomiopen ↗	1.1 min	59.7
22	Qwen3.7 PlusAlibaba	1.0 min	58.5
23	Mixtral 8x7BMistral AIopen	1.0 min	31.7
24	Qwen2.5 72BAlibabaopen ↗	58.6 s	42.2
25	North Mini CodeCohereopen	58.5 s	45.9
26	Gemma 3 27BGoogleopen ↗	58.5 s	38.9
27	Qwen3 VL 30B A3B InstructthinkingAlibabaopen ↗	57.6 s	44.0
28	Grok Build 0.1SpaceXAI	55.3 s	54.8
29	MiniMax M2.5MiniMaxopen ↗	55.2 s	53.6
30	GLM 5.3 FlashZhipu AIopen ↗	54.1 s	62.7
31	MiniMax-M2.1MiniMaxopen ↗	53.4 s	53.8
32	Grok 3 minithinkingSpaceXAI	53.0 s	49.4
33	Qwen2.5 Coder 32B InstructAlibabaopen ↗	51.6 s	42.1
34	Granite 3.3 8Bno reasoningIBMopen ↗	51.4 s	39.2
35	NewKimi K3.1 ProMoonshot AI	50.6 s	66.9
36	GPT-4 TurboOpenAI	48.5 s	40.9
37	GLM-4.6no reasoningZhipu AIopen ↗	48.4 s	46.7
38	LongCat 2.0Meituanopen ↗	48.3 s	49.4
39	GLM-4.5-AirZhipu AIopen ↗	48.2 s	46.6
40	MiMo-V2.5Xiaomiopen ↗	47.9 s	54.3
41	Granite 4.2 30BIBMopen ↗	47.6 s	48.7
42	Qwen3.8 MaxxhighAlibaba	46.7 s	57.1
43	Magistral Small 1.2Mistral AIopen	46.2 s	40.3
44	Gemini 4 ArgonGoogle	43.1 s	63.6
45	GLM 4.7no reasoningZhipu AIopen ↗	43.0 s	48.9
46	Gemma 2 27BGoogleopen ↗	42.9 s	41.0
47	GLM 5.3lowZhipu AIopen ↗	42.4 s	55.6
48	Phi-4Microsoftopen ↗	42.2 s	38.9
49	Qwen3 235B A22B 2507Alibabaopen ↗	41.6 s	50.4
50	Reka Flash 3Rekaopen ↗	41.2 s	38.9
51	Granite 4.2 8BIBMopen ↗	41.1 s	46.1
52	MiniMax M3MiniMaxopen ↗	40.7 s	56.4
53	Mistral Large 4Mistral AIopen	40.2 s	56.7
54	Qwen3 VL 235B A22B InstructthinkingAlibabaopen ↗	40.0 s	51.0
55	Claude Opus 4.6highAnthropic	39.7 s	62.7
56	Hermes 4 - Llama-3.1 405Bno reasoningNous Researchopen	39.6 s	43.4
57	Llama 3.1 70BMetaopen ↗	39.5 s	40.9
58	Qwen3 VL 8B InstructthinkingAlibabaopen ↗	39.4 s	42.7
59	Mistral Small 3Mistral AIopen ↗	39.2 s	39.3
60	MiniMax-M1MiniMaxopen	38.7 s	47.5

Show 100 more

Cite as: BenchLeader, “Time per task: code edit”, data as of 25 Oct 2026, https://www.benchleader.com/leaderboards/time-code-edit.