CONFIRMEDDate confidence: CONFIRMEDStated as firmHow firmly the source words the date: confirmed, scheduled, estimated, delayed or unverified.Show all CONFIRMED pins84% CONFIDENCEOverall confidence: 84%How well the pin's source and references back up its dates.Weighted average of how firmly 6 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better
MiniMax's[1] news post is dated 2025.12.23 and says the M2.1 API is now live on its Open Platform and the weights are open-source; OpenRouter[5] lists minimax/minimax-m2.1 on 23 December 2025 and Artificial Analysis[4] gives 23 December 2025, while the Hugging Face[6] repository was created on 20 December and its weights uploaded on 21 December.- Start
- Dec 23, 202590% CONFIDENCEfrom the source
MiniMax-M2.1 Released
MiniMax releases M2.1, an open-weight 230B-parameter mixture-of-experts coding and agent model that activates 10B parameters and scores 74.0 on SWE-bench Verified and 88.6 on its new VIBE full-stack benchmark.
- MiniMax released M2.1 on 23 December 2025, two months after M2, as an open-weight model for real-world coding and agent work: the API went live on the MiniMax Open Platform, MiniMax Agent is built on it and the weights are on Hugging Face under a modified MIT license[1][6]
- It targets multilingual software engineering, with MiniMax citing Rust, Java, Go, C++, Kotlin, Objective-C, TypeScript and JavaScript, plus native Android and iOS app development and web design quality, and says it "outperforms Claude Sonnet 4.5 and closely approaches Claude Opus 4.5" in multilingual scenarios[1][6]
- MiniMax's figures: 74.0 on SWE-bench Verified (M2 69.4, Claude Sonnet 4.5 77.2, GPT-5.2 thinking 80.0), 72.5 on SWE-bench Multilingual (M2 56.5), 49.4 on Multi-SWE-bench, 47.9 on Terminal-bench 2.0 (M2 30.0) and 88.6 on VIBE, the full-stack benchmark it open-sourced with the release (M2 67.5, Claude Sonnet 4.5 85.2, Claude Opus 4.5 90.7)[1][6]
- Artificial Analysis describes it as a 230B-parameter model with 10B active per token priced at $0.30 per million input tokens and $1.20 per million output tokens, and now scores it 21 on its Intelligence Index; OpenRouter lists the same price with a 204,800-token context[4][5]
- Kilo Code made M2.1 free for a limited time and reported nearly 200,000 Hugging Face downloads by 5 January 2026[3]
Notable features
- Interleaved thinking, which MiniMax calls one of the first open-source series to introduce it systematically, with more concise reasoning chains, faster responses and lower token use than M2[1]
- Works across coding agents and tools such as Claude Code, Droid, Cline, Kilo Code, Roo Code and BlackBox, and honors context files like Skill.md, Claude.md and agent.md and slash commands[1]
- A "Digital Employee" mode that operates a computer through text commands for office workflows, including Excel and Yahoo Finance tasks, and wider composite-instruction following[1]
- Served with SGLang, vLLM, Transformers and KTransformers; MiniMax recommends temperature 1.0, top_p 0.95 and top_k 40[1][6]
- The M2 series paper later reports the shared M2 base at 229.9B total and 9.8B active parameters[2]
Benchmarks
MiniMax's tables from the model card, merged; "-" means not reported[6].
| Benchmark | MiniMax-M2.1 | MiniMax-M2 | Claude Sonnet 4.5 | Claude Opus 4.5 | Gemini 3 Pro | GPT-5.2 (thinking) | DeepSeek V3.2 |
|---|---|---|---|---|---|---|---|
| SWE-bench Verified | 74.0 | 69.4 | 77.2 | 80.9 | 78.0 | 80.0 | 73.1 |
| Multi-SWE-bench | 49.4 | 36.2 | 44.3 | 50.0 | 42.7 | - | 37.4 |
| SWE-bench Multilingual | 72.5 | 56.5 | 68 | 77.5 | 65.0 | 72.0 | 70.2 |
| Terminal-bench 2.0 | 47.9 | 30.0 | 50.0 | 57.8 | 54.2 | 54.0 | 46.4 |
| SWT-bench | 69.3 | 32.8 | 69.5 | 80.2 | 79.7 | 80.7 | 62.0 |
| VIBE (average) | 88.6 | 67.5 | 85.2 | 90.7 | 82.4 | - | - |
| Toolathlon | 43.5 | 16.7 | 38.9 | 43.5 | 36.4 | 41.7 | 35.2 |
| BrowseComp | 47.4 | 44.0 | 19.6 | 37.0 | 37.8 | 65.8 | 51.4 |
| AIME25 | 83.0 | 78.0 | 88.0 | 91.0 | 96.0 | 98.0 | 92.0 |
| MMLU-Pro | 88.0 | 82.0 | 88.0 | 90.0 | 90.0 | 87.0 | 86.0 |
| GPQA-Diamond | 83.0 | 78.0 | 83.0 | 87.0 | 91.0 | 90.0 | 84.0 |
| HLE (no tools) | 22.2 | 12.5 | 17.3 | 28.4 | 37.2 | 31.4 | 22.2 |
| LiveCodeBench | 81.0 | 83.0 | 71.0 | 87.0 | 92.0 | 89.0 | 86.0 |
| IFBench | 70.0 | 72.0 | 57.0 | 58.0 | 70.0 | 75.0 | 61.0 |
| tau2-Bench Telecom | 87.0 | 87.0 | 78.0 | 90.0 | 87.0 | 85.0 | 91.0 |