BlogLLM Benchmark Scores as Published, Side by Side

MiniMax M2.7: Benchmark Scores as Published

Every result the catalog holds for MiniMax M2.7: 62 tests in 12 areas, from 5 pages. 33 printed by MiniMax itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.

Developer MiniMaxReleased 2026-03-18Input TextOutput TextWeights OpenData version 2026103001, as of 2026-10-30

Knowledge Work

Agentic Coding

TestValuePrinted bySetup as printed
SWE-bench Verified79.9%MiniMax, MiniMax M3checked 2026-10-15; printed on 1 more pageHarness: Claude Code
SWE-Bench Pro56.2%MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15; printed on 2 more pages-
Terminal-Bench 2.151.1%MiniMax, MiniMax M3checked 2026-10-15; printed on 1 more pageHarness: Terminus 2
SWE-bench Multilingual76.5%MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15; printed on 1 more page-
Terminal-Bench 2.057%MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15; printed on 1 more page-
NL2Repo-Bench39.8MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15; printed on 2 more pages-
SWE-Atlas-QnA (setting not stated)11.29%MiniMax, MiniMax M3checked 2026-10-15-
Multi-SWE-bench52.7%MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15-
SWE-Atlas-Test Writing18.89%MiniMax, MiniMax M3checked 2026-10-15-
SWE-fficiency13.98%MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code
VIBE-Pro55.6%MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15-
VIBE-V2 (MiniMax internal)37.89%MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code

ML Engineering

TestValuePrinted bySetup as printed
PaperBench30.6MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code
MLE-bench Lite (medal rate)66.6%MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15-
KernelBench Hard (TFLOPs relative to hardware peak)10.5%MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code
PostTrainBench (version not stated)13.1%MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code

Science and Math

TestValuePrinted bySetup as printed
GPQA Diamond87%Z.ai, GLM-5.1 model cardchecked 2026-10-15-
SciCode38.3%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15-
IMO-AnswerBench68.3%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15; printed on 1 more pageTools: none
75.1%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: with tools
AIME 202689.8%Z.ai, GLM-5.1 model cardchecked 2026-10-15-
HMMT November 202581%Z.ai, GLM-5.1 model cardchecked 2026-10-15-
HMMT February 202672.7%Z.ai, GLM-5.1 model cardchecked 2026-10-15-
CritPt0.6%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: none
MathArena Apex Shortlist51.9%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: with tools
28.9%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: none
GPQA86.6%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: none

Long Context

TestValuePrinted bySetup as printed
AA-LCR69.8%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15-
LOCA-Bench (256k environment description)0%MiniMax, MiniMax M3checked 2026-10-15-

General Knowledge and Reasoning

Code Generation

TestValuePrinted bySetup as printed
LiveCodeBench v677.2%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15-
LiveSQLBench (Base-Full v1)33.17%MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code
SVG-Bench (MiniMax internal)48%MiniMax, MiniMax M3checked 2026-10-15Harness: Claude Code

Tool Use and Workflows

Web Search and Research

TestValuePrinted bySetup as printed
BrowseComp76.3%MiniMax, MiniMax M3checked 2026-10-15; printed on 1 more pageTools: web browsing
DRACO66.77%MiniMax, MiniMax M3checked 2026-10-15-

Instruction Following

Multilingual

Preference and Composite Rankings

TestValuePrinted bySetup as printed
Artificial Analysis Intelligence Index50MiniMax, MiniMax-M2.7 model cardchecked 2026-10-15-

Price

API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked. Cost per task is measured by Vals AI.

PriceValuePrinted bySetup as printed
Cost per taskVals AI, USD, lower is better$2.82Vals AI, Vibe Code Bench v1.1 leaderboardchecked 2026-10-27Harness: OpenHands

More from MiniMax

Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.