MiMo-V2.6-Flash-RL review, pricing and limits

Flash-RL pairs a 1M-token context window with 128K max output and 59.2 tokens/second throughput, faster than its Pro-RL sibling. Choose it for high-volume, cost-sensitive agentic work; choose Pro-RL instead when the intelligence gap matters or when confirmed video and audio input is required.

MiMo-V2.6-Flash-RL is the efficiency-tier model in Xiaomi's MiMo-V2.6 series: a 309-billion-parameter Mixture-of-Experts system with 15 billion active parameters, MIT licensed since its September 2026 debut. Its Artificial Analysis Intelligence Index score of 38 sits well above the open-weight median, and it reads text and image input across a 1 million token context window at roughly a third of Pro-RL's per-token price.

Provider: Xiaomi Corporation · Family: MiMo V2.6

More about Xiaomi Corporation on HokAI

Context window: 1,000,000 tokens · Max output: 128,000

Input modalities: text, image · Output: text

About MiMo-V2.6-Flash-RL

MiMo-V2.6-Flash-RL is the efficiency-tier checkpoint of Xiaomi's MiMo-V2.6 series, released open-weight on September 21, 2026 under the MIT license, led by the same Luo Fuli-headed MiMo team as MiMo-V2.6-Pro-RL. It is a smaller sparse Mixture-of-Experts transformer: 309 billion total parameters with 15 billion active per token, routed across 256 experts (8 activated) over 48 layers. Xiaomi built it as the balance point of the two RL-trained V2.6 checkpoints, for high-frequency calls and large-scale workflows where trillion-parameter Pro-RL would cost too much.

Flash-RL scores 38 on the Artificial Analysis Intelligence Index, well above the open-weight median (18) though behind Pro-RL's 46.32. On agentic evals Xiaomi reports directly, it scores 87.6 on Terminal-Bench 2.1 and 80.8 on OSWorld-Verified, both close to Pro-RL's 89.9 and 82.0 and to the closed-source Claude Opus 5 (89.1, 83.4) named in the shared technical report. It trails badly on cybersecurity exploit evals: 25.3 on ExploitBench against Claude Opus 5's 70.0.

The model ships the same 1 million token context window as Pro-RL, with a 128,000 token max output and an identical 5-layer speculative decoder design. Artificial Analysis measured 59.2 output tokens per second, faster than Pro-RL's 48.6 but still under the open-weight class median of 85.7 t/s, plus a 3.80-second time-to-first-answer-token. Flash-RL was also the more verbose of the two on the Intelligence Index, generating 240 million output tokens against Pro-RL's 140 million.

Xiaomi's own product page and this model's Hugging Face card both list Flash-RL's input modality as text, image, video, and audio, matching Pro-RL. Artificial Analysis, which tests models directly rather than citing vendor claims, found only text and image input actually working through Flash-RL's API as of this listing. Output remains text only in every source.

Xiaomi's API prices Flash-RL far below Pro-RL: $0.14 per 1M input tokens on a cache miss, $0.0028 per 1M on a cache hit (about a 98% discount), and $0.28 per 1M output tokens. Artificial Analysis computes a blended rate of $0.06 per 1M tokens, a third of Pro-RL's $0.18.

Weights are published on Hugging Face and ModelScope under MIT. Self-hosting recipes are lighter than Pro-RL's: SGLang with tensor-parallel 8 and data-parallel 2, or vLLM with tensor-parallel-size 4, roughly half the GPU count Pro-RL calls for. Distribution is narrower than Pro-RL: only 1 API provider was live on Artificial Analysis's tracker at listing time, versus 3 for Pro-RL; access otherwise runs through the same first-party channels (Xiaomi's own API, AI Studio, MiMo Desktop, OpenRouter).

As with Pro-RL, Xiaomi's Trust Center had not published a MiMo-V2.6-specific training-data summary or system card at listing time. The shared technical report describes the same "Aligned RL" self-correction stage and adversarial/verifier hardening used to train both checkpoints together.

Flash-RL suits high-frequency, cost-sensitive workloads needing agentic and coding competence close to Pro-RL's, and teams that can accept its narrower API distribution or self-host on roughly half the GPU budget. It is weaker for tasks needing Pro-RL's top intelligence score, offensive-security work, or workflows relying on the vendor's claimed video/audio input, which Artificial Analysis could not confirm.

Flash-RL trained alongside Pro-RL in the same reinforcement-learning campaign: Xiaomi reports its run also completed 30 GRPO steps in under six days, sharing the roughly 750,000-trajectory batch structure, at a reported cost of $850,000, well below Pro-RL's $2.62 million.

There is no MiMo-V2.5-Flash predecessor in Xiaomi's comparison table, so Flash-RL's closest lineage comparison is MiMo-V2.5-Pro, which it beats on every shared benchmark, for example 87.6 vs 65.2 on Terminal-Bench 2.1. Xiaomi has already previewed MiMo-V3, reported to adopt a new "HySparse2" architecture to cut long-context serving costs, as the next model in the line. A smaller research sibling, MiMo-V2.6-Distill-Qwen-9B, was released alongside it for self-hosting on a single GPU.

Pricing

Verified 2026-09-28 on Xiaomi's own pricing page: fresh input runs $0.14 per 1M tokens, cache-hit input drops to $0.0028 per 1M (about 98% cheaper), and output runs $0.28 per 1M tokens, roughly a fifth of Pro-RL's rate across the board. A separate Token Plan subscription and batch API exist for high-volume or non-real-time work; those rates were not independently verified for this listing.

What a real job costs

Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.

Key Features

• Sparse Mixture-of-Experts backbone: 256 routed experts with 8 activated per token across 48 transformer layers and a 4096-dimension hidden size, about half the layer count of Pro-RL's 70-layer backbone.
• 1M-token context window: 128K max output tokens on the same 1M-token context as Pro-RL, paired with a 5-layer speculative decoder to keep long agentic sessions responsive.
• Vision-capable input: Reads text and image input; Xiaomi's own pages also list video and audio, though Artificial Analysis's independent testing did not confirm those two modalities working via the API.
• Groupwise agentic RL training: Trained in the same mixed reinforcement-learning run as Pro-RL (GRPO across coding, agent, visual, and cybersecurity tasks), using an automated grader that ranks passing rollouts by quality rather than a simple pass/fail signal.
• Lighter self-hosting footprint: Xiaomi's reference recipes need 4 GPUs (vLLM) or 8 GPUs (SGLang), roughly half of what the full Pro-RL checkpoint requires.

Pros

• Costs about a third of Pro-RL's blended per-token rate while staying within a few points of it on Terminal-Bench 2.1 and OSWorld-Verified.
• MIT license with no commercial-use restrictions, plus a lighter self-hosting footprint than Pro-RL (4-8 GPUs vs 8-16).
• Faster output than its larger sibling: 59.2 tokens/second against Pro-RL's 48.6, per Artificial Analysis.

Cons

• Vendor markets video and audio input identical to Pro-RL, but Artificial Analysis's independent tests only confirmed text and image working via the API.
• Hosted-API distribution is thinner than Pro-RL's right now: Artificial Analysis tracked just 1 provider for Flash-RL versus 3 for its sibling.
• Generated 240 million output tokens on the Intelligence Index, more verbose than the larger Pro-RL's 140 million.

Benchmarks

• OSWorld Verified: 80.8% vendor-reported · 28 Sep 2026 — Tasks completed by operating a real desktop, % solved.
• Terminal-Bench 2.1: 87.6% vendor-reported · 28 Sep 2026 — Multi-step tasks completed in a real command line, % solved.
• AA Intelligence Index: 38 cited: Artificial Analysis · 28 Sep 2026 — Composite of 10 evaluations run by Artificial Analysis, 0 to 100.
• AA blended price: $0.06/M cited: Artificial Analysis · 28 Sep 2026 — Price per 1M tokens at a 3:1 input to output blend, as listed by Artificial Analysis.
• Output speed: 59 tok/s cited: Artificial Analysis · 28 Sep 2026 — Median tokens written per second as measured by Artificial Analysis.

A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.

Visit MiMo-V2.6-Flash-RL Official Page