Fastest LLM (October 2026)
As of 21 Oct 2026, the fastest good LLM on independent benchmarks is K2 Horizon 375B A23B by MBZUAI (output speed 1115 tok/s), ahead of Grok 4.20 Multi-Agent (output speed 356 tok/s) and Muse Spark 1.2 (xhigh) (output speed 315 tok/s).
Output speed in tokens per second, among models that stream at least 100 tokens per second and score above 55 on the BenchLeader Index, so the list is fast models worth using rather than the fastest small ones.
40 models qualify · data as of 21 Oct 2026 · rebuilt once a day (the run starts at 22:03 UTC)
K2 Horizon 375B A23B #1 here: output speed 1115 tok/sBest value
Claude Haiku 5.5 xhighmost index points per dollar among the top 15 (61.5 at $0.200/M)| # | Model | Index | Price $/M | Speed tok/s |
|---|---|---|---|---|
| 1 | K2 Horizon 375B A23BMBZUAI | 56.3 | – | 1115 |
| 2 |  Grok 4.20 Multi-AgentSpaceXAI | 58.7 | $1.56 | 356 |
| 3 |  Muse Spark 1.2xhighMeta | 63.2 | $2.00 | 315 |
| 4 |  GPT-5.4 MinixhighOpenAI | 56.4 | $1.69 | 276 |
| 5 | Muse Spark 1.1Meta | 62.9 | $2.00 | 236 |
| 6 |  Gemini 3.6 FlashhighGoogle | 62.1 | $1.50 | 214 |
| 7 | Gemini 3 FlashthinkingGoogle | 59.2 | $1.13 | 214 |
| 8 |  DeepSeek V4.1 Flash 1017maxDeepSeek | 63.1 | $0.525 | 209 |
| 9 | Gemini 3.5 FlashhighGoogle | 63.2 | $3.38 | 208 |
| 10 | Deepseek v4 Flash VisionmaxDeepSeek | 58.3 | $0.660 | 202 |
| 11 |  Qwen3 7maxAlibaba | 60.7 | $3.75 | 201 |
| 12 | Claude Haiku 5.5xhighAnthropic | 61.5 | $0.200 | 192 |
| 13 | GPT-5-CodexhighOpenAI | 57.5 | $3.44 | 190 |
| 14 |  Quasar 438BmaxMultiverse Computing | 55.8 | $0.900 | 183 |
| 15 | Muse WondermaxMeta | 67.5 | $8.00 | 161 |
Full ranked list (40 models) · Best for voice agents, with the weighting stated · Methodology · the LLM benchmarks behind this ranking
Fastest LLM: questions
- What is the fastest good LLM in October 2026?
- As of 21 Oct 2026, the fastest good LLM on independent benchmarks is K2 Horizon 375B A23B by MBZUAI (output speed 1115 tok/s), ahead of Grok 4.20 Multi-Agent (output speed 356 tok/s) and Muse Spark 1.2 (xhigh) (output speed 315 tok/s). BenchLeader reconciles 229 benchmarks from 43 benchmark publishers and rebuilds the ranking once a day (the run starts at 22:03 UTC).
- Which is the best value?
- Claude Haiku 5.5 (xhigh) gives the most BenchLeader Index points per dollar among the top 15 here: 61.5 at $0.200 per million tokens blended, against an unpublished price for K2 Horizon 375B A23B.
- How is this ranked?
- Output speed in tokens per second, among models that stream at least 100 tokens per second and score above 55 on the BenchLeader Index, so the list is fast models worth using rather than the fastest small ones. Every score comes from an independent publisher and links to its source on the model page; self-reported launch numbers are excluded.
Cite as: BenchLeader, “Fastest LLM (October 2026)”, https://www.benchleader.com/best/fastest-llm, data as of 21 Oct 2026.