MiniMax
MiniMax M3
MiniMax-M3 is a multimodal foundation model from MiniMax. It supports text, image, and video inputs with text output, a 1M-token context window, and is suited for long-horizon agentic work, coding,...
Cost / 1M tokens
Low usage$0.30
input
$0.06
cached input
$1.20
output
Context Window
1M
Up to 512K max output
Modalities
text, image, video → text
Other tokenizer
Quickstart & API
minimax/minimax-m3Best for & Strengths
Key Strengths
Recommended Workloads in ChatOSS
- Coding agents
- Complex reasoning & debugging
- Screenshot-to-code workflows
Verified Evaluations
Benchmarks & Frontier Comparison
How MiniMax M3 compares against leading open-weight and proprietary frontier models across independent agentic coding and reasoning suites.
| Benchmark (Metric) | MiniMax M3 | DeepSeek V4.1 Flash | GLM 5.3 | Kimi K3.1 Pro | Claude Opus 5.5 |
|---|---|---|---|---|---|
| Agentic & Software Engineering | |||||
| ChatOSS Agentic Coding Index (%) | 22.2% | 63.2% | 58.3% | 65% | 74.1% |
| SWE-Bench Verified (%) | — | 91.5% | 95.4% | 94.8% | 97% |
| DeepSWE v1.1 (Pass@1 %) | 20.4% | 77% | 69% | 73.5% | 74.2% |
| Vibe Code Bench v1.1 (%) | 47.57% | 87.7% | 78.1% | 88.4% | 90.3% |
| Terminal-Bench 4.0 (%) | 1.01% | 42.3% | 42% | 46.2% | 65.2% |
| Code Migration (%) | 19.93% | 45.6% | 44.2% | 52% | 66.7% |
| Intelligence & Token Economics | |||||
| Artificial Analysis Intelligence Index (0–100) | 18.9 | 39.5 | 44.8 | 55.1 | 57.6 |
| Input Price (/ 1M tokens) | $0.30 | $0.15 | $1.40 | $2.00 | $4.00 |
| Output Price (/ 1M tokens) | $1.20 | $0.60 | $4.40 | $10.00 | $20.00 |
ChatOSS Agentic Coding Index
ChatOSS Benchmark Audit · 4-benchmark composite mean
DeepSWE v1.1
DeepSWE / Official Audit · Mini-SWE-Agent
Vibe Code Bench v1.1
Vals.ai · OpenHands
Terminal-Bench 4.0
Vals.ai / Artificial Analysis · Terminus 2
Code Migration
Vals.ai · Vals evaluation harness
Artificial Analysis Intelligence Index
Artificial Analysis
Design Arena: AGENTICGAMEDEV
Design Arena · Rank #18 (44.9% win)
Design Arena: ANDROIDNATIVE
Design Arena · Rank #25 (44.9% win)
Design Arena: BLENDER3D
Design Arena · Rank #15 (42.3% win)
Design Arena: CODETOVIDEO
Design Arena · Rank #6 (56.9% win)
Active Provider Endpoints (9)
Upstream inference providers serving MiniMax M3, with per-provider pricing, context limits, quantization, and 24-hour uptime.
| Provider | Input / 1M | Output / 1M | Cache Read / 1M | Context | Max Output | Quantization | Uptime (24h) |
|---|---|---|---|---|---|---|---|
DeepInfracache deepinfra/fp8/flex | $0.224 | $0.88 | $0.045 | 524K | 131K | fp8 | 65.5% |
CoreWeavecache coreweave/fp4 | $0.23 | $0.96 | $0.05 | 262K | 236K | fp4 | 99.7% |
GMICloudcache gmicloud/fp8 | $0.24 | $0.96 | $0.048 | 1M | 524K | fp8 | 99.7% |
DeepInfracache deepinfra/fp8 | $0.28 | $1.10 | $0.056 | 524K | 131K | fp8 | 97.6% |
StreamLakecache streamlake/fp8 | $0.30 | $1.20 | $0.06 | 1M | 512K | fp8 | 100.0% |
Venicecache venice/fp8 | $0.30 | $1.20 | $0.06 | 524K | 131K | fp8 | 93.0% |
Parasailcache parasail/fp8 | $0.30 | $1.20 | $0.06 | 1M | 524K | fp8 | 99.9% |
Minimaxcache minimax | $0.30 | $1.20 | $0.06 | 524K | 512K | — | 100.0% |
Parasailcache parasail/fast | $0.45 | $1.80 | $0.09 | 1M | 524K | fp8 | 99.5% |
Supported API Parameters
Parameters accepted by this model when called through ChatOSS or OpenRouter.
Deep-Dive Benchmarks & Teardowns
Independent benchmark analysis and real-world agent billing teardowns featuring MiniMax M3.
2026-10-29 · 11 min read
MiniMax M3.1 Flash may be the best open-source orchestrator agent right now
AutomationBench is the closest public test of orchestration, and MiniMax M3.1 Flash wins it on both the independent run (73% vs 67%) and the vendors' table (62.8 vs 60.5). It also leads HLE in Artificial Analysis's run, 48% to 43% — but DeepSeek V4.1 Flash 1017 still owns coding, cache-heavy bills and the MIT licence. The honest case, including the numbers that cut against it.
2026-10-27 · 10 min read
MiniMax M3.1 Flash vs. DeepSeek V4.1 Flash 1017: cheaper tokens, higher bill
Both list at $0.15/$0.60 per million tokens, and MiniMax wins the business-workflow benchmarks. But on vals.ai's Vibe Code Bench v1.1, DeepSeek's October 17 checkpoint scores 87.73% at $0.58 per test while M3.1 Flash scores 73.94% at $1.15 — because the cheaper token burns 66.86M of them per task. Here is the independent evidence and the cache-aware math.
Compare Similar Models
Browse all models →MiniMax
MiniMax M3.1 Flash
MiniMax-M3.1-Flash is a multimodal model from MiniMax. Compared with M3, it delivers significant improvements across multiple benchmarks, particularly in complex agentic tasks, coding, and cybersecurity. It is built for complex,...
MiniMax
MiniMax M2.7
MiniMax-M2.7 is a next-generation large language model designed for autonomous, real-world productivity and continuous improvement. Built to actively participate in its own evolution, M2.7 integrates advanced agentic capabilities through multi-agent...
MiniMax
MiniMax M2.5
MiniMax-M2.5 is a SOTA large language model designed for real-world productivity. Trained in a diverse range of complex real-world digital working environments, M2.5 builds upon the coding expertise of M2.1...
Anthropic
Claude Opus 5.5
Claude Opus 5.5 is Anthropic's flagship model for demanding reasoning, coding, and long-horizon agentic work, succeeding Claude Opus 5. It is particularly strong at multi-step changes in large codebases, code...
Moonshot AI
Kimi K3.1 Pro
Kimi K3.1 Pro is Moonshot AI's flagship open-weight multimodal reasoning model and a post-trained successor to Kimi K3, with a 1M-token context window. It is built for long-horizon coding, 3D...
Anthropic
Claude Fable 5.1
Claude Fable 5.1 improves on Claude Fable 5 across the board, with the biggest gains in agentic coding, long-running agentic workflows, and knowledge work: long code refactors, front-end and visual...