Skip to content
Bencher

Models · MiniMax · Out sinceReleased 18 Mar 2026

MiniMax M2.7

Available on OpenRouterOn OpenRouterAvailable nowGenerally availableCan run on your own serversOpen weights

MiniMax M2.7 is made by MiniMax. We don't have enough test results yet to rank it. It's cheap to use. Its makers have published it, so you can run it on your own servers.

Previous generation. M2.7-highspeed variant $0.60/$2.40. Context from OpenRouter.

Writing & creativity
—
Research & analysis
—
Coding
—
Price
Cheap$0.30 / $1.20
Price per 1M (blended)Blended / 1M
$0.52
MemoryContext
205K
Longest answerMax output
—
Test resultsResults
17 (2 independent2 indep.)
Out sinceReleased
18 Mar 2026
Made byVendor
MiniMax
UnderstandsInputs
text

Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.

Thinking levels

Reasoning settings

Default

Read more

Links

Using it through OpenRouter

On OpenRouter

OpenRouter is a service that gives access to many AI models in one place. This model is listed there as minimax/minimax-m2.7.

Route it as minimax/minimax-m2.7 at $0.21 in / $0.84 out per 1M tokens, 205K context. Listed since 18 Mar 2026.

frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p

Run it yourself

Self-hosting

On your own servers,Licence, sizeunder your own control.and hardware.

Compare all open models →All open-weights models →

Details comingselfHost data pending

This model can be downloaded and run on your own servers. We're collecting its licence, size and hardware needs now.Open-weights model; licence, parameter count, hardware tier and base-model availability will appear after the next data run.

Every test result

Every result

All the numbers,with where they came from.

Every result we've found for this model, with who measured it and a link to where we read it.

All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.

All results for this model
NotesNotes
BrowseComp76.3%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗1 Jun 2026—MiniMax M2.7 column in the M3 benchmark table image (GitHub README)
GDPval-AA (v1)1495Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—
KernelBench Hard10.5%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗1 Jun 2026Claude CodeMiniMax M2.7 column in the M3 benchmark table image (GitHub README)
LLM Creative Story-Writing Benchmark (Lech Mazur)-3.8DefaultMiniMax-M2.7Independent testIndependentLech Mazur (LLM Creative Story-Writing Benchmark) ↗28 Sep 2026—rank 54/56; Thurstone comparison score (centered at 0); est. win chance 7%; 95% bootstrap -3.898 to -3.622
MCP Atlas49.4%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗1 Jun 2026—MiniMax M2.7 column in the M3 benchmark table image (GitHub README)
MLE-Bench Lite66.6%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—Medal rate over 22 ML competitions
Multi-SWE-bench52.7%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—
NL2Repo-Bench39.8%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—
SWE Atlas - Codebase QnA11.3%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗1 Jun 2026mini-SWE-agentMiniMax M2.7 column in the M3 benchmark table image (GitHub README)
SWE-bench Multilingual76.5%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—
SWE-Bench Pro (public, v1)56.2%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—
SWE-bench Verified79.9%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗1 Jun 2026Claude CodeMiniMax M2.7 column in the M3 benchmark table image (GitHub README)
Terminal-Bench 2.057.0%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—Terminal Bench 2; harness not stated in MiniMax card
Terminal-Bench 2.151.1%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗1 Jun 2026Terminus 2MiniMax M2.7 column in the M3 benchmark table image (GitHub README)
Toolathlon46.3%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—Reported as 'Toolathon'
Vectara Hallucination Leaderboard (HHEM)12.9%DefaultMiniMaxAI/minimax-m2p7Independent testIndependentVectara ↗22 Sep 2026—factual consistency 87.1 %; answer rate 99.4 %; avg summary 131.9 words; HHEM-2.3 judge; effort not stated (API default)
VIBE-Pro (MiniMax)55.6%Defaultsetting not statedMaker's own figureVendor-reportedMiniMax ↗18 Mar 2026—In-house