SShogo Bench

MiniMax-M3

MiniMax multimodal model for long-context coding, perception, and agent planning

Shogo Agent Index

Estimated#61 of 92
36.7range 26.5–46.8 · 20% of weight measured
Agentic coding
37
estimated
Long-horizon autonomy
49
30% measured
Knowledge-work agents
31
32% measured
Computer use
24
60% measured
Security & context learning
37
estimated

Price

Input
$0.30/M
Output
$1.20/M
Cache read
$0.060/M
3:1 blend
$0.52/M
Agent-effective
$0.15/M

Cheapest host for agents: DeepInfra at $0.11/M.

Agent-effective = 20 input tokens per output token, 85% cache hits. Why.

Agentic benchmarks

BenchmarkScoreNormalisedIndex weightHarness / source
SWE-bench Verifiedest.8%predicted 69
Terminal-Bench 2.0est.8%predicted 44
DeepSWEest.6%predicted 30
FrontierSWEest.5%predicted 6
CursorBenchest.4%predicted 29
GSO-Benchest.4%predicted 12
METR Time Horizonsest.10%predicted 67
Vending-Bench 2$2,1586%source
PostTrainBenchest.4%predicted 19
APEX-Agents37.7%8%source
GDPvalest.7%predicted 35
Remote Labor Indexest.5%predicted 2
DeepResearch Benchest.5%predicted 46
OSWorld 2.04.6%6%source
OSWorldest.4%predicted 54
Cybenchest.5%predicted 55
CL-benchest.5%predicted 18

Other benchmarks

BenchmarkCategoryScoreSource
Chess PuzzlesReasoning17.0%source
CritPtReasoning3.7%source
DTBenchKnowledge78.9%source
FrontierCodeCoding14.7%mini-swe-agent
GPQA DiamondReasoning90.9%source
LMCAKnowledge33.7%source
OTIS Mock AIMEMath71.1%source
ProofBenchMath18.0%source
SciCodeCoding47.1%source
SimpleBenchReasoning45.8%source
WebDev ArenaCoding1487source

API hosts

Live prices and uptime for every host serving MiniMax-M3 via OpenRouter.

HostInputOutputCache readAgent $/MUptime 24hQuant
DeepInfra$0.22$0.88$0.045$0.11100.0%fp8
CoreWeave$0.23$0.96$0.050$0.1296.9%fp4
GMICloud$0.24$0.96$0.048$0.12100.0%fp8
DeepInfra$0.28$1.10$0.056$0.14100.0%fp8
Minimax$0.30$1.20$0.060$0.15100.0%—
Parasail$0.30$1.20$0.060$0.1599.9%fp8
Venice$0.30$1.20$0.060$0.1599.3%fp8
Parasail$0.45$1.80$0.090$0.2299.6%fp8
Featherless$0.55$2.20$0.060$0.2310.3%fp8