VECTOR WIREAI INTELLIGENCE
UTC
Refresh Models Deals Regulatory Sources

Gemma 4 26B A4B vs MiMo V2.5 Pro

39 SHARED BENCHMARKS

Across 39 shared benchmarks, Gemma 4 26B A4B scores higher on 10 and MiMo V2.5 Pro on 29. The widest gap is SWE-bench Verified, where MiMo V2.5 Pro scores 78.9 against 17.4. Gemma 4 26B A4B is the cheaper of the two on tracked API pricing ($0.13 against $0.43 per million input tokens).

GOOGLEVSXIAOMI39 SHARED1029 HEAD-TO-HEADNEWEST SCORE ADDED

At a glance

MakerGoogleXiaomi
Released
Price per 1M tokens input / output$0.13 / $0.40$0.43 / $0.87
Cost of 1M in + 1M out$0.53 2.5× less$1.30
Head-to-head of 39 shared benchmarks10 wins29 wins
Scores tracked independently verified135 14101 4

Release dates per Artificial Analysis. Prices: Artificial Analysis for Gemma 4 26B A4B; Artificial Analysis for MiMo V2.5 Pro. ◆ = independently verified score.

Where each leadsby capability area · benchmark wins

AreaGemma 4 26B A4BWINSMiMo V2.5 Pro
Reasoning21Gemma 4 26B A4B leads 2 of 3 · widest: Humanity's Last Exam 19.3 vs 14.8
Coding16MiMo V2.5 Pro leads 6 of 7 · widest: SWE-bench Verified 17.4 vs 78.9
Agentic12MiMo V2.5 Pro leads 2 of 3 · widest: AA IT-Bench SRE 23.6 vs 38.2
Factuality11Even, 1–1 of 2
Instruction Following10Gemma 4 26B A4B leads 1 of 1 · widest: IFBench 72.4 vs 42.7
Long Context10Gemma 4 26B A4B leads 1 of 1 · widest: AA-LCR 65.7 vs 41.7
Math02MiMo V2.5 Pro leads 2 of 2 · widest: AIME 2026 88.3 vs 93.6
Multimodal12MiMo V2.5 Pro leads 2 of 3 · widest: CharXiv (RQ) 69 vs 81

Biggest gaps

Gemma 4 26B A4B pulls furthest ahead on

  1. LiveCodeBench v677.1 vs 39.6
  2. IFBench72.4 vs 42.7
  3. AA-LCR65.7 vs 41.7

MiMo V2.5 Pro pulls furthest ahead on

  1. SWE-bench Verified78.9 vs 17.4
  2. SWE-bench Pro57.2 vs 13.8
  3. Terminal-Bench 2.165.2 vs 39

Every shared benchmark39 · grouped by area

Reasoning 3

Instruction Following 1

Long Context 1

AA-LCR65.741.7

Math 2

Multimodal 3

Each score is the one the model's own page shows — the most authoritative tracked result for that benchmark, on the benchmark's standard methodology (source-attributed). ↓ marks lower-is-better metrics; ◆ an independently verified score. Open either model for its full surface, provenance and pricing. Quoted rates are the price-setter row we currently track for each model — its direct or vendor-official listing where one exists (direct, direct), otherwise the lowest tracked offer.

Compare Gemma 4 26B A4B withALL PAIRINGS →

Compare MiMo V2.5 Pro withALL PAIRINGS →