Venice Atlas
← Back to model catalog

VENICE ATLAS · MODEL DOSSIER · MINIMAX M3 PREVIEW

Model overview

MiniMax M3 Preview

MiniMax-M3 preview is a 1.4T-parameter frontier model from MiniMax for coding, agentic workflows, and complex reasoning, served at fp8 with a 512K context window.

minimax-m3-preview · venice.ai
Text · Venice catalog$0.30 per 1M tokensPrice on Venice AI
ModalityTextOn Venice AI
Primary price$0.30 per 1M tokensper 1M tokens
Context window524KMax tokens
Benchmarks35Benchmark results

Benchmark results

See the tests behind each score. They use the benchmark’s original scoring scale.

Frontend Development#828.65 ratingHealthcare#2029.39 ratingMultimodal#3428.23 ratingProductivity#810.45 ratingSearch#1722.24 ratingStructured Output#429.16 ratingSystems#612.82 rating
Agents8 results
BankerToolBenchbankertoolbench · agents
76.1%
ReportedOct 2026
DRACOdraco · agents
73.2%
ReportedOct 2026
Finance Agent v2finance-agent-v2 · agents
48.3%
ReportedOct 2026
LOCA-Bench (256k)loca-bench-256k · agents
49.3%
ReportedOct 2026
OfficeQA Proofficeqa-pro · agents
45.1%
ReportedOct 2026
PostTrainBenchposttrainbench · agents
37.1%
ReportedOct 2026
SpreadSheetBench-v1spreadsheetbench-v1 · agents
89.3%
ReportedOct 2026
YC-Benchyc-bench · agents
21.0%
ReportedOct 2026
Coding11 results
CL-benchcl-bench · code
20.5%
ReportedOct 2026
Claw-Evalclaw-eval · code
74.5%
ReportedOct 2026
KernelBench Hardkernelbench-hard · code
28.8%
ReportedOct 2026
LiveSQLBenchlivesqlbench · code
40.2%
ReportedOct 2026
NL2Reponl2repo · code
42.1%
ReportedOct 2026
PaperBenchpaperbench · code
52.6%
ReportedOct 2026
SVG-Benchsvg-bench · code
63.7%
ReportedOct 2026
SWE Atlas - Codebase QnAswe-atlas-codebase-qna · code
37.9%
ReportedOct 2026
SWE Atlas - Test Writingswe-atlas-test-writing · code
30.8%
ReportedOct 2026
SWE-fficiencyswe-fficiency · code
34.8%
ReportedOct 2026
VIBE-V2vibe-v2 · code
50.1%
ReportedOct 2026
General intelligence1 result
GDPval-Rubricsgdpval-rubrics · general
74.8%
ReportedAug 2026
Math3 results
IMO 2025imo-2025 · math
83.3%
ReportedOct 2026
LiveBenchlivebench · math
70.0%
ReportedOct 2026
USAMO 2026usamo-2026 · math
85.7%
ReportedOct 2026
Multimodal3 results
OSWorld-Verifiedosworld-verified · multimodal
70.1%
ReportedOct 2026
Video-MMEvideo-mme · multimodal
85.4%
ReportedOct 2026
VideoMMMUvideommmu · multimodal
84.6%
ReportedOct 2026
Reasoning4 results
APEX-Agentsapex-agents · reasoning
27.7%
ReportedOct 2026
BrowseCompbrowsecomp · reasoning
83.5%
ReportedOct 2026
FrontierCode 1.1frontiercode-1.1 · reasoning
14.7%
ReportedOct 2026
MCP Atlasmcp-atlas · reasoning
74.2%
ReportedOct 2026

Learn what each benchmark measures ↗

Related models

More from minimax

minimax-m25MiniMax M2.5$0.27 input per 1M tokens ↗minimax-m27MiniMax M2.7$0.38 input per 1M tokens ↗minimax-m3-1-flashMiniMax M3.1 Flash$0.19 input per 1M tokens ↗