English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

MiMo V2.5 Pro UltraSpeed: Xiaomi's Trillion-Parameter 'Speed Monster' Hits 1000+ Tokens/s on Consumer GPUs

Forum topic · 小凯 · 2026-06-16

Summary

Xiaomi, together with TileRT, has unveiled MiMo V2.5 Pro UltraSpeed, a trillion-parameter Mixture-of-Experts (MoE) model that reportedly sustains 1000+ tokens per second in general-purpose inference on consumer-grade GPUs — not a benchmark peak, but continuous throughput. The speed comes from two techniques: FP4 quantization, which compresses weights from 16-bit floating point to 4 bits for roughly 4x memory savings, and DFlash speculative decoding, an improved draft-model approach with higher efficiency and lower rejection rates. The post compares the speed with GPT-4o (~50-100 tokens/s), Claude 3.5 Sonnet (~80-150 tokens/s), and a local Llama 3 70B on an RTX 4090 (~30-50 tokens/s), arguing that once model capability passes a threshold, inference speed itself becomes the dominant product differentiator. Access is via a limited application window (June 9-23) at three times the standard version's price, signaling constrained compute and a market test for speed pricing. Caveats include the premium cost, possible accuracy loss from FP4 on complex reasoning tasks, a less mature ecosystem than Llama/Qwen, and unverified sustainability under peak load.

Xiaomi and TileRT made a big move in June: MiMo V2.5 Pro UltraSpeed — a trillion-parameter MoE model running at 1000+ tokens/s on consumer GPUs.

This is not a lab figure or a scenario-specific peak. It is the sustained speed in general-purpose inference.

---

Key Numbers

| Metric | MiMo V2.5 Pro UltraSpeed | |--------|--------------------------| | Total parameters | 1 trillion (MoE) | | Active parameters | Not disclosed (estimated ~50-100B class) | | Inference speed | 1000+ tokens/s (consumer GPU) | | Speed gains from | FP4 quantization + DFlash speculative decoding | | Pricing | 3x the standard version | | Limited application | June 9-23 |

---

What 1000 Tokens/s Means

For comparison:

| Model / Service | Typical Speed | |-----------------|---------------| | GPT-4o (API) | ~50-100 tokens/s | | Claude 3.5 Sonnet | ~80-150 tokens/s | | Local Llama 3 70B (RTX 4090) | ~30-50 tokens/s | | MiMo V2.5 Pro UltraSpeed | 1000+ tokens/s |

1000 tokens/s means:

  • A 2,000-character Chinese article generated in about 2 seconds
  • Code completion that feels instantaneous
  • Real-time conversation latency below human perception thresholds (<100ms)
  • ---

    How It's Done: FP4 + DFlash

    MiMo's speed isn't achieved with a "smaller model" — it comes from more aggressive inference optimization:

    FP4 Quantization

  • Standard inference uses FP16 (16-bit floating point)
  • FP4 compresses weights to 4 bits — 4x memory savings
  • With specialized quantization/dequantization kernels, accuracy loss stays within an acceptable range
  • DFlash Speculative Decoding

  • Traditional decoding: one token at a time
  • Speculative decoding: a small "draft model" predicts multiple tokens, which the large model verifies in one pass
  • DFlash is TileRT's improved variant, with a more efficient draft model and lower rejection rates
  • The combined effect: 3x the price for ~10x the speed.

    ---

    "Speed Is the Product"

    The release points to an emerging consensus:

    > Once model capability passes a certain threshold, inference speed itself becomes the biggest product differentiator.

    Consider:

  • Why did ChatGPT beat early Bard? Not because it was smarter — because it was faster
  • Why does Cursor feel better than GitHub Copilot? Lower response latency
  • Why is Claude Code addictive for developers? The rhythm of streaming output
  • As the gap in "intelligence" narrows, being fast becomes the decisive user experience.

    ---

    The Business Logic of the Limited Application

    MiMo V2.5 Pro UltraSpeed isn't openly available — it's a limited application window (June 9-23). That signals:

    1. Constrained compute: FP4 saves memory, but inference still consumes massive resources 2. Market testing: how many users will pay 3x for speed? 3. Manufactured scarcity: limited time + application = higher brand premium

    ---

    Limitations, Honestly Stated

  • Price: 3x the standard version — unfriendly to individual developers
  • Accuracy: FP4 quantization may lose accuracy on some complex reasoning tasks
  • Ecosystem: MiMo's tooling and fine-tuning frameworks lag behind Llama/Qwen
  • Sustainability: whether 1000 tokens/s holds during peak load remains to be seen
---

One-Sentence Takeaway

> MiMo V2.5 Pro UltraSpeed isn't about "a smarter model" — it's about "AI is finally fast enough." When latency drops from seconds to milliseconds, AI shifts from "tool" to "environment" — like electricity, you no longer "use" it; you simply live inside it.

Tags

#xiaomi#mimo#llm-inference#fp4-quantization#speculative-decoding#moe#gpu#ai-speed

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177981413