MiniMax-M3
MiniMax · released
MiniMax's flagship pay-as-you-go model: natively multimodal with a 1M-token context, aimed at coding agents and long office or research workflows at a low per-token price.
Best for: Low-cost coding agents · Multimodal input · Long documents
- Intelligence Index
- 29.2
- #26 of 38
- Price per 1M tokens
- $0.30 in $1.2 out
- Blended $0.52 at 3:1
- Context window
- 1M
- Output speed
- 94 tokens/s
- $0.51 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall40 · #27 of 38
Based on 3 of 3 benchmarks
- Coding23 · #27 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work36 · #29 of 40
Based on 6 of 6 benchmarks
- Reasoning & science34 · #27 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability65 · #19 of 39
Based on 2 of 2 benchmarks
- Vision59 · #20 of 31
Based on 2 of 2 benchmarks
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#26 of 38· Artificial Analysis
- Vals Index36.5%
#23 of 26· $3.08 per test· Vals AI
- Arena · Text1440 ±4
#26 of 33· 58,757 votes· Arena
Coding
#26 of 38· Artificial Analysis
- SciCode47.1%
#28 of 38· Artificial Analysis
- Arena · WebDev1482 ±6
#26 of 34· 18,086 votes· Arena
- Vibe Code Bench47.6%
#23 of 26· Vals AI
- Code Migration20.9%
#22 of 26· Vals AI
Agents & knowledge work
- GDPval-AA37.3
#25 of 39· Artificial Analysis
- Arena · Agent-6.9 ±0.9
#25 of 29· Arena
- Finance Agent v248.3%
#23 of 26· Vals AI
#24 of 26· Vals AI
- Legal Research Bench29.8%
#23 of 26· Vals AI
- Tax Agent Bench49.7%
#23 of 26· Vals AI
Reasoning & science
- Humanity's Last Exam39.0%
#21 of 39· Artificial Analysis
- CritPt3.7%
#30 of 39· Artificial Analysis
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA Diamondsaturating92.9%
#7 of 27· Artificial Analysis
Knowledge & reliability
#20 of 39· Artificial Analysis
- AA-LCR83.0%
#9 of 39· Artificial Analysis
Vision
- MMMU-Pro78.6%
#13 of 27· Artificial Analysis
- Arena · Vision1237 ±7
#17 of 19· 18,308 votes· Arena
Reported by MiniMax
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- SWE-bench Pro · Claude Code scaffold59%
- SWE-bench Verified · Claude Code scaffold80.5%
- MCP Atlas74.2%
Specifications
- Lab
- MiniMax
- Released
- Jun 1, 2026
- Availability
- Generally available
- Context window
- 1,000,000 tokens
- Max output
- —
- Accepts
- Text, Images, Video
- Knowledge cutoff
- Not published
- Weights
- Open
- Licence
- MiniMax Community License
- Parameters
- ~428B total · ~23B active
- Reasoning control
- Thinking on, adaptive or off
- API model ID
MiniMax-M3
Pricing
- Input / 1M tokens
- $0.30
- Cached input / 1M
- $0.06
- Output / 1M tokens
- $1.2
- Blended (3:1)
- $0.52
- Measured cost / task
- $0.51
Worth knowing
- Weights are downloadable, but the community licence attaches conditions to commercial use.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.