MiniMax M2.5
Open-weight coding contender hitting ~80% SWE-bench Verified — proof the open tier has closed much of the engineering gap.
When to use it
Open coding agents and competitive self-hosted SWE workflows.
Watch-outs
Broader Arena/AA coverage still thinner than the biggest labs.
Benchmarks
Human preference ranking from blind pairwise chats. Higher is better; top frontier models cluster within ~50–80 Elo.
Artificial Analysis composite across agents, coding, science, and general evaluations (v4.1 weighting).
Percent of real GitHub issues resolved end-to-end. Strong signal for agentic coding usefulness.
PhD-level science questions. Separates frontier reasoning models better than saturated knowledge tests.
Harder multi-choice knowledge/reasoning suite than classic MMLU.
Frontier closed-ended academic difficulty across many domains.