Quick Take
Released in late April 2026, both models represent the latest in AI development. MiMo-V2.5 by Xiaomi establishes a clear lead in core performance metrics, while xAI’s Grok 4.3 (low) offers a competitive alternative with a distinct performance profile.
Benchmark Read
MiMo-V2.5 leads in the primary intelligence and coding indices, scoring 49 and 42.1 respectively, compared to Grok 4.3 (low)’s 43.9 and 31.6. In specific benchmarks, MiMo-V2.5 shows superior results in HLE (0.252 vs 0.173), SciCode (0.431 vs 0.419), TerminalBench Hard (0.416 vs 0.265), and TAU2 (0.906 vs 0.888). Grok 4.3 (low) holds a slight advantage in IFBench (0.809 vs 0.671) and LCR (0.64 vs 0.626). Math index data remains unavailable for both models.
Cost and Speed
MiMo-V2.5 is significantly more cost-effective, with a blended price of $0.72/1M tokens compared to Grok 4.3 (low)’s $1.56/1M. Xiaomi’s model also delivers faster output speeds at 91.485 tok/s versus 80.284 tok/s. Most notably, MiMo-V2.5 offers a time to first token of 2.699s, drastically outperforming Grok 4.3 (low)’s 13.852s, which may impact real-time application responsiveness.
Best Fit
Grok 4.3 (low) is best suited for users who prioritize performance on specific instruction-following tasks (IFBench) and LCR benchmarks. MiMo-V2.5 is the ideal choice for developers and enterprises seeking high-speed, cost-efficient coding assistance and superior general intelligence performance.
Comments (0)
to join the discussion
No comments yet
Be the first to share your thoughts!