Mimo 2.6

reported speed:

23.6 tokens/s generation · 37.5 tokens/s prompt processing

quant:

MQ-IQ2-XXS-XS-Q8 (GGUF)

kv:

q8_0

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo V2.6 Flash MOPD mixed GGUF at 23.59 t/s decode and 37.54 t/s prefill on an AMD Strix Halo 128GB. Setup is llama.cpp (Vulkan) with MQ-IQ2-XXS-XS-Q8 GGUF and q8_0 KV cache, 256K context, thinking disabled, one request at a time. Longer inputs slow decode to 15.89 t/s at 4,860 tokens and 7.39 t/s at 48,600 tokens. MTP was disabled; a preliminary test showed MTP 3 taking 225.591 s versus 39.896 s without it. An agent benchmark scored 330/900 (36.7%) with all ten tasks timing out.

reported speed:

49.6 tokens/s generation · 2370.2 tokens/s prompt processing

quant:

2.20 bpw (EXL3)

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo-V2.6-Flash-RL at 49.57 tok/s decode (p50, single stream) on an NVIDIA RTX 6000 Ada 96GB. Setup is ExLlamaV3 (vcruz305 fork) with a 2.20 bpw EXL3 pack (86.94 GB) at 65,536-token context and a 65,536-token KV pool; prefill measured 2,370.2 tok/s p50. With the DFlash drafter attached, decode rises to 184.11 tok/s p50 and prefill is 2,257.7 tok/s. Aggregate throughput under load reaches 236.2 tok/s at C=8 without the drafter and 330.6 tok/s with it; the eight-stream per-stream decode p50 is 34.95 and 54.46 tok/s respectively. The 2.50 bpw pack (98.48 GB) does not fit a 96 GB card. On a DGX Spark / GB10 unified-memory host the 2.50 bpw pack runs at 34.9 tok/s decode (code) and 23.7 tok/s (prose) with draft_accept 0.78.

reported speed:

5.6 tokens/s generation

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo-V2.6-Flash running a 262144-token prompt end to end on four ranks, with a decode step at 8 tokens of context going from 3.63 to 5.63 tokens a second. Setup is a four-rank run of the model's own code, with a 256k prompt reaching 48.32 tok/s and a 64k prompt at chunk 4096 reaching 104.4 tok/s (95.5 at chunk 2048). The gains come from a fixed 2^26-score row step in the attention block loop and from cutting 145 cudaStreamSynchronize calls a decode step; a free-memory budget instead of the constant measured 8246.7 s at 256k against 5424.9 s for the constant.

reported speed:

53.3 tokens/s generation

quant:

MXFP4 (MXFP4)

kv:

fp8

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo-V2.6-Flash-RL at 53.31 tok/s per stream at C1 on two DGX Sparks (GB10, 121.7 GiB unified memory each) with vLLM tensor parallel 2 and DFlash speculative decoding (7 draft tokens). Setup is vLLM with MXFP4 experts, fp8 KV cache, 300K max context, marlin MoE backend, GPU memory utilization 0.90, max-num-seqs 8, KV pool 1,835,052 tokens. Aggregate throughput is 155.77 tok/s at six streams; cold prefill ranges from 1,946.6 tok/s at 2K to 656.4 tok/s at 248K; TTFT 0.367 s at C1. User notes tool-call storms in agent use and that prose/narrative decode is slow due to low DFlash acceptance.

reported speed:

109.7 tokens/s generation · 5123.0 tokens/s prompt processing

quant:

MXFP4 (MXFP4)

kv:

FP8

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo-V2.6-Flash-RL at 109.7 tok/s decode on one stream on one RTX 5090 plus four DGX Spark (GB10) systems. Setup is the mimo26f-afd engine with MXFP4 weights and FP8 KV cache, attention-FFN disaggregation over RoCE v2 RDMA, DFlash speculative decoding, up to 1,048,576 tokens of context. Decode reaches 279.1 tok/s across six streams and 416.7 tok/s across sixteen; cold prefill is 3,630 / 5,123 / 4,816 / 4,283 tok/s at 2K / 8K / 32K / 64K. The comparison is a 4-Spark vLLM TP4 reference without the 5090, so the uplift includes the fifth device.

reported speed:

246.0 tokens/s generation

kv:

FP8

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo-V2.6-Flash-RL at 246 tok/s decode at C1 (90.0 decode steps/s) on two RTX PRO 6000 Blackwell GPUs. Setup is vLLM with FP8 KV cache, TP2 at 0.985 utilization, DFlash drafter with 7 tokens, video off, 1.31M token KV cache. Decode is 42.3 steps/s at C8 and 30.6 at C16; 246 tok/s at C1 on mixed real tasks. Prefill is 9.7K / 9.1K tok/s at 8K / 32K. First start autotunes b12x for about 15 minutes.

reported speed:

26.1 tokens/s generation · 310.0 tokens/s prompt processing

quant:

IQ2_M (GGUF)

kv:

q8_0

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo 2.6 Flash-RL at 26.1 tok/s decode (tg128) and ~310 tok/s prefill (pp4096) on a Strix Halo 128GB machine. Setup is llama.cpp (Vulkan, Radeon 8060S) with a custom IQ2_M-class GGUF at 2.76 bpw and q8_0 KV cache, 32768 context, -ub 2048. Prefill drops to 193 tok/s at the default -ub 512. The 100.4 GiB quant is measured against the native MXFP4 GGUF: KLD 0.164 mean, 88.2% same top-1 token, PPL ratio 1.128. MTP self-speculation is included in the file but does not speed up decode (draft acceptance ~50%), so speculation is off.

reported speed:

41.6 tokens/s generation · 1718.3 tokens/s prompt processing

quant:

Q8_0 (GGUF)

kv:

q4_0

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo 2.6 Distill Qwen 9B at 41.63 t/s generation and 1718.29 t/s prompt processing on an RTX 5060 Ti 16GB. Setup is llama.cpp (Llama UI) with GGUF Q8_0 weights, q4_0 KV cache, 122880-token context, and Flash Attention enabled. The run produced 11979 output tokens over 4 min 47 s from a 1929-token prompt; the first generated version failed and a second review pass by the model produced the final working demo.

reported speed:

68.3 tokens/s generation

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo V2.6 Pro RL at 68.3 t/s on eight DGX Spark units, up from 17.8 t/s without speculation. Setup is vLLM with DFlash speculative decoding on official weights; the figure includes prefill and request overhead. Outputs matched byte for byte across 12 cases, including two baseline mistakes. Four concurrent requests and a 257K-token input were also tested.

reported speed:

42-51 tokens/s generation

kv:

fp8

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports MiMo-V2.6-Flash-RL at ~42–51 t/s on 2× DGX Spark (GB10) with vLLM, TP=2 over RoCE, fp8 KV cache, full 1,048,576 context, and DFlash with 7 draft tokens. Setup uses the tonyd2wild recipe's sm121-v11-dflash2 image. Decode is ~42–51 t/s on agentic code at default sampling, 69 t/s on the recipe's coding benchmark at temperature 0, and ~20–23 t/s on prose. Deep prefill is the weak spot at ~160 t/s past 800K, so a cold 1M prompt takes ~67 min. User documents three serving bugs: empty streaming replies with thinking on (fixed by pre-opening the thinking tag in the chat template), dropped earlier reasoning in tool loops (fixed by template and chat_utils.py fallbacks), and a hidden 2,048-token output cap from generation_config.json (fixed with --override-generation-config max_new_tokens 131072). A 1M needle test retrieved 3/3 hidden codes from a 996K-token prompt.

reported speed:

49.1 tokens/s generation · 1238.4 tokens/s prompt processing

mtp (multi-token prediction):

on

Reported by the source; GPU count, offloading and concurrent requests can change this figure. Check the full setup before comparing.

User reports Mimo 2.6 Flash at 49.1 t/s generation and 1238.4 t/s prompt processing on an M5 Ultra 256GB, averaged over five trials at 32,000 prompt tokens and 64 generation tokens. Setup is a custom mlx-vlm patch loading the original weights, with MTP enabled but apparently not working. A second run at 64,000 prompt tokens and 1024 generation tokens averaged 38.1 t/s generation and 981.9 t/s prompt processing. User also compares Qwen3.8 Flash-Next FP8 on oMLX, reaching 39.7 to 43.9 t/s generation across 32,768 to 200,000 token prompts.