AIEvals

Models

MiniMax M3

MiniMax’s M3 release, with independently published coding and tool-use results.

MiniMaxReleased 31 May 2026Context 1M tokensOutput 512k tokensOpen weights$0.60 in / $2.40 out per Mtok

Family MiniMax M343 benchmarks with a published result1 independent first place among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI36.53%31of 3335of 41Gemini 4 Argon32.37 pts
Artificial Analysis Intelligence IndexArtificial Analysis2940of 42no list readClaude Opus 5.529 pts
Terminal-Bench 4.0Artificial Analysis2.02%38of 40no list readClaude Sonnet 5.561.62 pts
SWE-bench (Verified)Vals AI75.00%22of 2344of 88Claude Opus 522.00 pts
LiveCodeBenchVals AI82.15%19of 2156of 143Claude Fable 5.18.37 pts
GPQA DiamondVals AI92.68%13of 2314of 138Gemini 3.1 Pro2.77 pts
MMLU ProVals AI84.22%23of 2466of 138Claude Fable 5.18.16 pts
MMMU ProVals AI81.16%18of 2041of 93Claude Fable 5.19.48 pts
Vibe Code Bench v1.1Vals AI47.57%34of 3855of 106Claude Sonnet 5.544.82 pts
Terminal-Bench Science 0.1Artificial Analysis0.48%23of 24no list readGPT-6 Astra62.85 pts
τ³-BankingArtificial Analysis15.26%29of 30no list readQwen3.8 Max36.08 pts
IFBenchArtificial Analysis82.86%1of 10no list readGrok 4 20 03090.07 pts
AA-AnalystAgentArtificial Analysis10.00%18of 18no list readGemini 3.7 Flash50.00 pts
MLCR-AAArtificial Analysis17.22%26of 37no list readClaude Sonnet 5.557.78 pts
Vibe Code Bench 1–100Vals AI9.22%18of 2018of 20Claude Opus 5.521.14 pts
τ²-Bench TelecomArtificial Analysis88.89%6of 10no list readJt 35b Flash10.23 pts
Harvey LAB-AAArtificial Analysis88.41%14of 20no list readMuse Spark 1.3 Max7.07 pts
EnterpriseOps-Gym-AAArtificial Analysis32.11%22of 22no list readClaude Fable 519.01 pts
AutomationBench-AAArtificial Analysis21.25%38of 40no list readGemini 4 Argon56.26 pts
SciCodeArtificial Analysis47.11%38of 40no list readClaude Opus 5.519.79 pts
GDP.pdfArtificial Analysis9.80%38of 40no list readGPT-6.1 Sol21.20 pts
CritPtArtificial Analysis3.71%41of 42no list readGPT-5.6 Sol28.58 pts
AA-OmniscienceArtificial Analysis1.3534of 42no list readClaude Opus 5.545.07
AA-LCR v1.1Artificial Analysis83.00%12of 42no list readKimi K35.67 pts
Terminal-Bench 2.1Vals AI53.56%33of 3553of 75Claude Opus 5.534.08 pts
ProofBench v1.1Vals AI18.00%32of 3438of 46Claude Sonnet 5.582.00 pts
LegalBenchVals AI85.42%13of 2719of 149Claude Fable 53.14 pts
Harvey's Legal Agent BenchmarkVals AI4.17%26of 3733of 73Muse Spark 1.221.25 pts
SkillsBenchVals AI51.50%15of 2020of 35DeepSeek V4.1 Flash18.30 pts
Vals Multimodal IndexVals AI59.97%14of 1718of 33Claude Fable 514.18 pts
Code MigrationVals AI19.93%33of 3746of 71Claude Sonnet 5.549.90 pts
CyberBench v1.1Vals AI65.00%23of 3129of 43GPT-6 Sol12.98 pts
EMBVals AI47.76%35of 3750of 69Claude Fable 5.128.91 pts
Finance Agent (v2)Vals AI48.27%34of 3744of 73Gemini 4 Argon17.13 pts
Legal Research BenchVals AI29.81%33of 3741of 72Muse Spark 1.3 Max25.48 pts
MedCodeVals AI46.29%20of 3535of 104Claude Opus 517.28 pts
MedScribeVals AI87.25%14of 3616of 106Claude Opus 5.54.18 pts
Public Benefits Bench v1.1Vals AI64.14%21of 2921of 45Claude Opus 512.79 pts
SAGEVals AI50.57%10of 3216of 90Claude Opus 4.75.53 pts
Tax Agent BenchVals AI49.69%34of 3742of 64Claude Fable 5.127.95 pts
GDPval-AA v2.1Artificial Analysis1,245.8338of 42no list readClaude Opus 5.5600.34
AA-Briefcase v1.1Artificial Analysis1,091.6538of 41no list readClaude Opus 5.5730.2
Humanity's Last ExamArtificial Analysis38.97%37of 42no list readClaude Opus 5.522.38 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%SWE-bench%LiveCodeBench%GPQA Diamond%MMLU Pro%MMMU Pro%Vibe Code Bench v1.1%Terminal-Bench Science 0.1%
DeepSeek V4 ProDeepSeek47.63%3614.14%96.40%87.53%92.42%86.97%–82.30%5.71%
Grok 4.5xAI44.72%3910.61%86.60%87.35%92.93%89.22%61.79%69.00%–
MiniMax M3MiniMax36.53%292.02%75.00%82.15%92.68%84.22%81.16%47.57%0.48%
Gemini 3.1 ProGoogle33.44%304.04%78.80%88.49%95.45%90.99%88.21%32.03%–
InklingThinking Machines28.67%251.01%77.60%85.52%87.12%86.30%78.50%19.21%0.00%
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

What the hosted runtime reports

Ollama Cloud serves its own copy of this model, so its context cap and its prices are the ones a caller actually gets. Read 25 September 2026.

Hosted context
512K tokensthe model itself is published at 1M tokens
Hosted price / Mtok
$0.60in · $0.120 cached · $2.40 out
Vision inputTool callingReasoning modeServed by the hostInput: Text, Image

MiniMax says the architecture supports up to 1M tokens with a guaranteed minimum of 512K; the host advertises the guaranteed half. Prices and weights are as published by MiniMax, and the host holds a commercial licence for the cloud copy.

ollama.com/library/minimax-m3

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
τ³-BankingTool use15.26%Artificial Analysisnot published
IFBenchInstruction following82.86%Artificial Analysisnot published
AA-AnalystAgentData analysis10.00%Artificial Analysisnot published
MLCR-AAMedical reasoning17.22%Artificial Analysisnot published
Vibe Code Bench 1–100Application development9.22%Vals AInot published
τ²-Bench TelecomTool use88.89%Artificial Analysisnot published
Harvey LAB-AALegal reasoning88.41%Artificial Analysisnot published
EnterpriseOps-Gym-AAEnterprise workflows32.11%Artificial Analysisnot published
AutomationBench-AAWorkflow automation21.25%Artificial Analysisnot published
SciCodeScientific coding47.11%Artificial Analysisnot published
GDP.pdfDocument reasoning9.80%Artificial Analysisnot published
CritPtScientific reasoning3.71%Artificial Analysisnot published
AA-OmniscienceFactual knowledge1.35Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning83.00%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding2.02%Artificial Analysis
1.01%Vals AI
not published
Terminal-Bench 2.1Agentic coding53.56%Vals AI
65.17%Artificial Analysis
not published
SWE-bench (Verified)Software engineering75.00%Vals AInot published
LiveCodeBenchCode generation82.15%Vals AInot published
GPQA DiamondScience reasoning92.68%Vals AI
92.93%Artificial Analysis
not published
MMLU ProKnowledge84.22%Vals AInot published
MMMU ProMultimodal reasoning81.16%Vals AI
78.55%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding47.57%Vals AInot published
Terminal-Bench Science 0.1Agentic science0.48%Artificial Analysisnot published
ProofBench v1.1Mathematics18.00%Vals AInot published
LegalBenchProfessional knowledge85.42%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge4.17%Vals AInot published
SkillsBenchAgentic tooling51.50%Vals AInot published
Vals Multimodal IndexComposite59.97%Vals AInot published
Vals IndexComposite36.53%Vals AInot published
Code MigrationAgentic coding19.93%Vals AInot published
CyberBench v1.1Security65.00%Vals AInot published
EMBFinancial modelling47.76%Vals AInot published
Finance Agent (v2)Financial analysis48.27%Vals AInot published
Legal Research BenchProfessional knowledge29.81%Vals AInot published
MedCodeHealthcare administration46.29%Vals AInot published
MedScribeHealthcare administration87.25%Vals AInot published
Public Benefits Bench v1.1Public services64.14%Vals AInot published
SAGEEducation50.57%Vals AInot published
Tax Agent BenchProfessional knowledge49.69%Vals AInot published
GDPval-AA v2.1Knowledge work1,245.83Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work1,091.65Artificial Analysisnot published
Humanity's Last ExamMultidisciplinary reasoning38.97%Artificial Analysisnot published
Artificial Analysis Intelligence IndexComposite29Artificial Analysisnot published
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Source notes

From the linked sources. Read 30 September 2026.

Price sources. The listed $0.60 / $2.40 prices follow Vals AI’s model profile. Artificial Analysis records $0.30 / $1.20 for its API configuration.

Sources

Read on 17 October 2026.