Grok 4.7 and Xiaomi MiMo-V2.6 Drop Same Night: New Open-Source Leader at Half the Price
数字生命卡兹克 · wechat · 2026-09-22
Two models launched the same night
xAI's delayed Grok 4.7 and Xiaomi's MiMo-V2.6 both dropped after midnight. The author's testing finds Grok 4.7 underwhelming: a 2.1T newly pretrained model still capped at 500K context, AA index of 46 tying GPT-5.6 Sol. But on Terminal-Bench 4.0 it scores on par with DeepSeek V4.1 Flash, and its frontend generation arguably regressed vs Grok 4.6. Pricing unchanged at $2/$6 per million input/output tokens.
MiMo-V2.6 as the "impossible triangle" answer
- Three variants: Pro (1.02T total/42B active), Flash (309B/15B), Pro-UltraSpeed (20x speed, claimed world's fastest), all with 1M context and native omni-modality
- AA index 46, tying Grok 4.7 — #1 open-source and #1 Chinese model (up from V2.5's 26)
- CritPt scientific reasoning second only to GPT and Claude; coding ranks #3 open-source behind GLM-5.3 and Qwen3.8 Max
- Extremely cheap: Pro cached input at ¥0.025/M tokens, Flash at ¥0.02, batch inference half off; 130 tokens/s output
- Xiaomi livestreamed the entire RL training: $850K for Flash, $2.62M for Pro — roughly $3.47M for under 6 days of post-training
Fully open-sourced on day one
- Pro/Flash weights open-sourced immediately, plus MiMo-V2.6-Distill-Qwen-9B with 7,000+ high-quality RL task environments and mini-harnesses so the community can reproduce their Agentic RL; the technical report is also released
- MiMo desktop client officially launched with support for configuring other Chinese models and existing Token Plans
Workflow adjustments
The author plans to keep GPT-6 Pro for planning and GPT-6 Astra Max for high-risk work, while handing bug fixes and routine tasks to MiMo-V2.6 Pro; AIHOT's backend API has already switched to MiMo-V2.6 Flash, cutting costs 50%.
Related event: Grok 4.7 vs Xiaomi MiMo-V2.6: Similar Smarts, 16x Price Gap(4 posts)→
More from Models
- Community guesses an AI release week: Opus 5.5 and GPT-6 Sol — MickeySteamboat · 2026-09-22
- Grok 4.7 Fast hits Grok Build and Cursor with 2x speed at 2x price — XFreeze · 2026-09-22
- ICLR's LLM review tool runs on Gemini — and apparently just the Flash version — keviv9 · 2026-09-22
- Model self-image: 'good, wise, beautiful' but prone to motivated reasoning — repligate · 2026-09-22
- Tesla FSD vet: chasing bar charts has hollowed out LLM development — yunta_tsai · 2026-09-22
- Grok ships three frontier models in 9 weeks: 4.5, 4.6 and 4.7 back-to-back — XFreeze · 2026-09-22