Xiaomi MiMo 2.6 Pro hands-on: 2x cheaper than DeepSeek V4, refuses no cybersecurity queries
deedydas · x · 2026-09-22
Deedydas tried Xiaomi's newly released MiMo 2.6 Pro, which claims to be the best open-source model:
- Extremely cheap: with standard agentic-coding/cache-hit assumptions, roughly 15x cheaper than Kimi K3, 6x cheaper than GLM 5.3, 2x cheaper than DeepSeek V4; $0.0036/M cache hit, $0.435/M input, $0.87/M output.
- Pareto frontier: unquestionably on the price/performance Pareto frontier.
- Strong on cybersecurity: refuses almost no security queries — asking about finding buffer overflows in ImageMagick yields detailed output; scores 95 on CyberGym.
- Ultraspeed mode: claimed up to 20x speed, in practice 3x throughput on OpenRouter, 150tps p50, at 10x the net price.
- Solid multimodal: built-in TTS for informational videos, auto-added speech/sound effects, and surprisingly good music composition in a DAW.
The technical report goes deep into how they scaled RL. Whether it's truly the best open model is too early to call, but it's a strong contender with many use cases, especially in cyber.
More from Models
- YC-backed Monte launches post-training platform to build proprietary models from your agents' real work — ycombinator · 2026-11-12
- Leak Claims OpenAI Will Launch Entire GPT-6.2 Family Tomorrow to Counter Opus 5.5 and Fable 5.2 — heypearlai · 2026-11-12
- ARC-AGI-3 leaderboard: models score 90%+ but at $10K+ per task — haider1 · 2026-11-12
- LMArena Launches Agent Arena for Head-to-Head Agent Evaluation — arena · 2026-11-12
- LLMs lack hyperlocal municipal data, explains disputed Texas power outage debate — csuwildcat · 2026-11-12
- Claude API's per-message effort lets you switch reasoning levels without breaking prompt cache — lyraxana · 2026-11-12