AIEvals

Models

Mistral Medium 3.5

Mistral AI’s Medium 3.5 release, with independently published reasoning and tool-use results.

Mistral AIReleased 29 April 2026Context 262k tokensOutput 80k tokensOpen weights$1.50 in / $7.50 out per Mtok

Family Mistral Medium 3.535 benchmarks with a published resultNo first place on the independent columns 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
Artificial Analysis Intelligence IndexArtificial Analysis1442of 42no list readClaude Opus 5.544 pts
Terminal-Bench 4.0Artificial Analysis0.00%40of 40no list readClaude Sonnet 5.563.64 pts
SWE-bench (Verified)Vals AI66.40%23of 2372of 88Claude Opus 530.60 pts
GPQA DiamondVals AI34.85%23of 23134of 138Gemini 3.1 Pro60.60 pts
MMLU ProVals AI75.33%24of 24111of 138Claude Fable 5.117.05 pts
Vibe Code Bench v1.1Vals AI2.89%38of 38100of 106Claude Sonnet 5.589.50 pts
τ³-BankingArtificial Analysis15.05%30of 30no list readQwen3.8 Max36.29 pts
IFBenchArtificial Analysis68.78%7of 10no list readGrok 4 20 030914.15 pts
AA-AnalystAgentArtificial Analysis12.50%17of 18no list readGemini 3.7 Flash47.50 pts
MLCR-AAArtificial Analysis1.67%36of 37no list readClaude Sonnet 5.573.33 pts
τ²-Bench TelecomArtificial Analysis94.15%4of 10no list readJt 35b Flash4.97 pts
Harvey LAB-AAArtificial Analysis69.10%19of 20no list readMuse Spark 1.3 Max26.38 pts
EnterpriseOps-Gym-AAArtificial Analysis33.72%20of 22no list readClaude Fable 517.40 pts
AutomationBench-AAArtificial Analysis6.30%39of 40no list readGemini 4 Argon71.21 pts
SciCodeArtificial Analysis40.16%40of 40no list readClaude Opus 5.526.74 pts
GDP.pdfArtificial Analysis2.80%40of 40no list readGPT-6.1 Sol28.20 pts
CritPtArtificial Analysis0.00%42of 42no list readGPT-5.6 Sol32.29 pts
AA-OmniscienceArtificial Analysis-36.842of 42no list readClaude Opus 5.583.22
AA-LCR v1.1Artificial Analysis69.33%42of 42no list readKimi K319.34 pts
Terminal-Bench 2.1Vals AI38.95%35of 3569of 75Claude Opus 5.548.69 pts
ProofBench v1.1Vals AI9.00%33of 3442of 46Claude Sonnet 5.591.00 pts
Harvey's Legal Agent BenchmarkVals AI0.42%36of 3754of 73Muse Spark 1.225.00 pts
Vals Multimodal IndexVals AI34.77%17of 1733of 33Claude Fable 539.38 pts
Code MigrationVals AI5.13%37of 3764of 71Claude Sonnet 5.564.70 pts
EMBVals AI11.19%37of 3767of 69Claude Fable 5.165.48 pts
Finance Agent (v2)Vals AI32.10%37of 3763of 73Gemini 4 Argon33.30 pts
Legal Research BenchVals AI9.13%37of 3764of 72Muse Spark 1.3 Max46.16 pts
MedCodeVals AI33.75%35of 3582of 104Claude Opus 529.82 pts
MedScribeVals AI67.73%36of 3696of 106Claude Opus 5.523.70 pts
SAGEVals AI37.61%28of 3264of 90Claude Opus 4.718.49 pts
Tax Agent BenchVals AI27.29%37of 3759of 64Claude Fable 5.150.35 pts
GDPval-AA v2.1Artificial Analysis747.142of 42no list readClaude Opus 5.51,099.07
AA-Briefcase v1.1Artificial Analysis520.840of 41no list readClaude Opus 5.51,301.05
Humanity's Last ExamArtificial Analysis13.76%42of 42no list readClaude Opus 5.547.59 pts

Nearby on Artificial Analysis Intelligence Index

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

ModelArtificial Analysis Intelligence IndexindexTerminal-Bench 4.0%SWE-bench%GPQA Diamond%MMLU Pro%Vibe Code Bench v1.1%
MiniMax M3MiniMax292.02%75.00%92.68%84.22%47.57%
InklingThinking Machines251.01%77.60%87.12%86.30%19.21%
Mistral Medium 3.5Mistral AI140.00%66.40%34.85%75.33%2.89%
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 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.

Parameters
128B dense parameters
Hosted context
256K tokensthe model itself is published at 262k tokens
Vision inputTool callingReasoning modeInput: Text, Image

Runs locally from an 80GB download rather than through a hosted endpoint, so there is no per-token price; Mistral releases the weights under a modified MIT licence.

ollama.com/library/mistral-medium-3.5

Results by benchmark

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

BenchmarkIndependent runAs reported by a publisher
τ³-BankingTool use15.05%Artificial Analysisnot published
IFBenchInstruction following68.78%Artificial Analysisnot published
AA-AnalystAgentData analysis12.50%Artificial Analysisnot published
MLCR-AAMedical reasoning1.67%Artificial Analysisnot published
τ²-Bench TelecomTool use94.15%Artificial Analysisnot published
Harvey LAB-AALegal reasoning69.10%Artificial Analysisnot published
EnterpriseOps-Gym-AAEnterprise workflows33.72%Artificial Analysisnot published
AutomationBench-AAWorkflow automation6.30%Artificial Analysisnot published
SciCodeScientific coding40.16%Artificial Analysisnot published
GDP.pdfDocument reasoning2.80%Artificial Analysisnot published
CritPtScientific reasoning0.00%Artificial Analysisnot published
AA-OmniscienceFactual knowledge-36.8Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning69.33%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding0.00%Artificial Analysisnot published
Terminal-Bench 2.1Agentic coding38.95%Vals AI
50.56%Artificial Analysis
not published
SWE-bench (Verified)Software engineering66.40%Vals AInot published
GPQA DiamondScience reasoning34.85%Vals AI
74.85%Artificial Analysis
not published
MMLU ProKnowledge75.33%Vals AInot published
MMMU ProMultimodal reasoning
64.86%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding2.89%Vals AInot published
ProofBench v1.1Mathematics9.00%Vals AInot published
Harvey's Legal Agent BenchmarkProfessional knowledge0.42%Vals AInot published
Vals Multimodal IndexComposite34.77%Vals AInot published
Code MigrationAgentic coding5.13%Vals AInot published
EMBFinancial modelling11.19%Vals AInot published
Finance Agent (v2)Financial analysis32.10%Vals AInot published
Legal Research BenchProfessional knowledge9.13%Vals AInot published
MedCodeHealthcare administration33.75%Vals AInot published
MedScribeHealthcare administration67.73%Vals AInot published
SAGEEducation37.61%Vals AInot published
Tax Agent BenchProfessional knowledge27.29%Vals AInot published
GDPval-AA v2.1Knowledge work747.1Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work520.8Artificial Analysisnot published
Humanity's Last ExamMultidisciplinary reasoning13.76%Artificial Analysisnot published
Artificial Analysis Intelligence IndexComposite14Artificial 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.

Sources

Read on 17 October 2026.