MiniMax M2.5
MiniMax M2.5 is capable in long context; and behind the leaders in instruction following, coding, agentic tasks, reasoning, and factuality. Too few results yet to rate safety, math, multimodal tasks, or multilingual tasks.
Capability profile
Bars show the model's median result as a share of the leading model's, per capability. Select a row to see its results.
Too few results yet to rate Safety, Math, Multimodal or Multilingual.
Price
$0.30input$1.20outputper million tokens
From Artificial Analysis · 3 providers tracked · All prices
Evidence
40results on33benchmarks
- 11 independently verified
- 15 aggregator
- 14 vendor-reported
From 9 sources · latest Oct 8, 2026 · How verification works
API features
Tool callingStructured outputsReasoning
As listed by OpenRouter
Research
14 papers reference MiniMax M2.5MiniMax M2.5 benchmark results
40 results on 33 benchmarks, grouped by capability. Each bar is the result as a share of the capability leader's; every result links to its source.
21.3% behind the leader1 of 3 ranked benchmarks measured
- 73.33Oct 8, 2026
27.2% behind the leader1 of 3 ranked benchmarks measured
- IFBench71.63Oct 8, 2026aa_ifbench
Show 1 more instruction following resultHide 1 instruction following result
- 70.00Aug 24, 2026
29.4% behind the leader5 of 10 ranked benchmarks measured
- 80.20Oct 7, 2026
- SciCode42.59Sep 4, 2026aa_scicode
- 55.40Oct 7, 2026
- 1386.69May 22, 2026
- Terminal-Bench Hard34.85Oct 8, 2026aa_terminalbench_hard
Show 5 more coding resultsHide 5 coding results
- 44.40Aug 24, 2026
- 68.30May 1, 2026
- 75.80Sep 11, 2026
- SWE-bench Verified76.10Jun 5, 2026SWE-Bench Verified (OpenCode)
- SWE-bench Verified79.70Jun 5, 2026SWE-Bench Verified (Droid)
35.1% behind the leader2 of 7 ranked benchmarks measured
- 76.30Oct 7, 2026
- 33.91Jun 15, 2026
35.8% behind the leader4 of 6 ranked benchmarks measured
- GPQA Diamond84.85Oct 8, 2026gpqa
- Humanity's Last Exam20.53Oct 8, 2026aa_hle
- 4.86May 10, 2026
- 1.14Oct 8, 2026
Show 2 more reasoning resultsHide 2 reasoning results
- GPQA Diamond85.20Aug 24, 2026GPQA-D
- Humanity's Last Exam19.40Aug 24, 2026HLE w/o tools
36.0% behind the leader3 of 4 ranked benchmarks measured
- 9.10May 2, 2026
- AA-Omniscience · Accuracy26.18Oct 8, 2026omniscienceAccuracy
- AA-Omniscience · Non-hallucination11.88Oct 8, 2026omniscienceNonHallucination
More results
Benchmarks outside the capability baskets. They are not ranked against a leader.
- 6.33Oct 8, 2026
- 75.80Aug 29, 2026
- 63.67May 10, 2026
- vectara_avg_summary_length137.20May 2, 2026Average Summary Length (Words)
- vectara_answer_rate98.20May 2, 2026Answer Rate
- vectara_factual_consistency90.90May 2, 2026Factual Consistency Rate
Show 10 more resultsHide 10 results
- 55.60Jun 18, 2026
- AA Intelligence22.80Oct 8, 2026aa_intelligence_index
- -38.87Oct 8, 2026
- AIME25 no tools86.30Aug 24, 2026AIME25
- 37.43Jun 18, 2026
- 79.70Jun 12, 2026
- 59.00Jun 5, 2026
- 51.30Oct 7, 2026
- 76.10Jun 12, 2026
- τ²-Bench Telecom (AA run)95.32Oct 8, 2026aa_tau2
MiniMax M2.5: common questions
Who makes MiniMax M2.5?
MiniMax M2.5 is made by MiniMax.
When was MiniMax M2.5 released?
MiniMax M2.5 was released on Feb 12, 2026, according to Artificial Analysis.
What is MiniMax M2.5 good at?
MiniMax M2.5 is capable in long context; and behind the leaders in instruction following, coding, agentic tasks, reasoning, and factuality. Too few results yet to rate safety, math, multimodal tasks, or multilingual tasks.
How much does MiniMax M2.5 cost?
MiniMax M2.5 costs $0.30 per million input tokens and $1.20 per million output tokens, according to Artificial Analysis. We track its price at 3 providers. At a mix of three input tokens to one output token, it is cheaper than 59% of the 330 priced models we track.
How many benchmarks has MiniMax M2.5 been tested on?
We track 40 results for MiniMax M2.5 on 33 benchmarks from 9 sources, 11 of them independently verified. The latest was recorded on Oct 8, 2026.
Which API features does MiniMax M2.5 support?
OpenRouter lists tool calling, structured outputs, and reasoning for MiniMax M2.5.
About this record
Where MiniMax M2.5's numbers come from, and every name it appears under.
- Tracked since
- Apr 27, 2026
- Newest source mention
- Aug 10, 2026
Where the results come from
Verification: 40 scores · 11 independently verified · 15 aggregator-attributed · 14 vendor-reported. How these tiers are assigned
From 9 sources on 8 sites. Artificial Analysis supplies 15 of them; the 11 independently verified results come from 5 sites. Bars are coloured by trust tier.
- artificialanalysis.ai15
- huggingface.co10
- api.llm-stats.com4
- raw.githubusercontent.com4
- swebench.com3
- arcprize.org2
- datasets-server.huggingface.co1
- labs.scale.com1