Xiaomi Open-Sources MiMo-V2.6, Scoring 46 on AA Index, Just 1 Point Behind GPT-5.6
On 09-24 Xiaomi released and open-sourced the MiMo-V2.6 family (Pro and Flash), two natively multimodal models built around scaled reinforcement learning. Pro scored 46 on the Artificial Analysis Intelligence Index, topping all open-weight and Chinese models, just 1 point behind GPT-5.6 Sol (max) at 47, while its weighted average cost is only about $0.13 per task—roughly 1/15 of GPT-5.6's. The gap between open-source and top closed frontier models narrowed to under one point with an order-of-magnitude cost advantage, with direct implications for model pricing and the open-source ecosystem.
Confirmed
- MiMo-V2.6 includes Pro and Flash, both natively multimodal, emphasizing scaled reinforcement learning
- Pro scored 46, first among open-weight and Chinese models, ahead of Qwen3.8-Max, Kimi K3, GLM-5.3 (@FellMentKE) and DeepSeek V4.1 Flash at 39 (@buffering112)
- GPT-5.6 Sol (max) scored 47; Pro's weighted per-task cost is $0.13, about 1/15 of GPT-5.6's (@kimmonismus)
- @buffering112 reports user-measured costs 15-20% lower than DeepSeek
- Pro's agent-benchmark performance is said to be on par with Claude Opus 5 and GPT-5.6 Sol
Why it matters
- @davidyin44 notes this continues the trend of Chinese vendors offering near-frontier closed-model capability at very low cost; Xiaomi, a phone maker, entering open-source models intensifies competition
- A 1-point gap with roughly an order-of-magnitude cost difference could disrupt closed-model pricing logic and boost the practical value of open-source models
2026-09-24 ~ 2026-09-24 · 7 related posts
- Episode 1: Xiaomi MiMo Finishes RL Training, DeepSWE Score Jumps to 72.57(2026-09-21, 2 posts)
- Episode 2: Xiaomi open-sources MiMo-V2.6, topping open-weight leaderboard at 46 with livestreamed RL training(2026-09-22, 47 posts)
- Episode 3: Xiaomi's MiMo-V2.6 Pro tested: big gains in coding, 3D games and multimodal(2026-09-22, 2 posts)
- Episode 4: Xiaomi Open-Sources MiMo-V2.6, Scoring 46 on AA Index, Just 1 Point Behind GPT-5.6(2026-09-24, 7 posts)
Primary sources
- Xiaomi ships open-weights MiMo-V2.6-Pro, tops open-source AI index at 46, rivaling Claude Opus 5 — ycombinator · 2026-09-24
- Xiaomi's MiMo V2.6 Pro Tops Open-Source at $0.13 per Task, One Point Behind GPT-5.6 — FellMentKE · 2026-09-24
- [source] Xiaomi's MiMo-V2.6-Pro tops open-weight models: 46 on AA Index, 15x cheaper than GPT-5.6 — kimmonismus · 2026-09-24
- [source] Xiaomi's open model sits 1 point behind GPT-5.6 at ~1/15th the cost per task — davidyin44 · 2026-09-24
- [source] Xiaomi's Open-Weight MiMo V2.6 Pro Cuts Agent Costs 15-20% vs DeepSeek — buffering_112 · 2026-09-24
2 near-duplicate retellings: kimmonismus · SucceededMind