Quick answer: MiniMax M2.5 is MiniMax's January 2026 model scoring 76.3% BrowseComp, 80.2% SWE-Bench Verified, and 76.8% BFCL-v3 Multi-turn. The upgrade to M2 — substantially higher BrowseComp.
Where MiniMax M2.5 leads
Where it lags
Best for: BrowseComp web research pipelines; agentic multi-turn tool use (BFCL multi-turn); SWE-Bench software engineering.
MiniMax M2.5 is MiniMax's January 2026 successor to M2. The 76.3% BrowseComp is dramatically higher than M2's 44.0% — MiniMax's major focus for the M2.5 upgrade. The 80.2% SWE-Bench Verified and 76.8% BFCL-v3 Multi-turn further demonstrate strong agentic capabilities.
| Field | Value |
|---|---|
| Organization | MiniMax |
| License | MiniMax Commercial License |
| Release date | January 2026 |
| Modality | Text only |
Available via MiniMax API. Refer to MiniMax pricing.
| Benchmark | Score | Source | Date |
|---|---|---|---|
| BrowseComp | 76.3% | Benchgen evaluation | 2026-01 |
| SWE-Bench Verified | 80.2% | Benchgen evaluation | 2026-01 |
| BFCL-v3 Multi-turn | 76.8% | Benchgen evaluation | 2026-01 |
| Model | BrowseComp | SWE-Bench Verified | License |
|---|---|---|---|
| MiniMax M2.5 | 76.3% | 80.2% | MiniMax Commercial |
| MiniMax M2 | 44.0% | — | MiniMax Commercial |
| Kimi K2.6 | 83.2% | — | Apache 2.0 |
M2.5 vs M2: BrowseComp jumps from 44.0% to 76.3%. For open-source BrowseComp: Kimi K2.6 (83.2%).
Specs from MiniMax's M2.5 release (January 2026) and Benchgen evaluations. Last updated 2026-07-24.
This model isn’t on any benchmark leaderboard yet.