MiniMax-M2.7 is our first model deeply participating in its own evolution. M2.7 is capable of building complex agent harnesses and completing highly elaborate productivity tasks, leveraging Agent Teams, complex Skills, and dynamic tool search. For more details, see our blog post.
M2.7 initiates a cycle of model self-evolution: during development, we let the model update its own memory, build dozens of complex skills for RL experiments, and improve its own learning process based on experiment results. An internal version of M2.7 autonomously optimized a programming scaffold over 100+ rounds — analyzing failure trajectories, modifying code, running evaluations, and deciding to keep or revert — achieving a 30% performance improvement. On MLE Bench Lite (22 ML competitions), M2.7 achieved a 66.6% medal rate, second only to Opus-4.6 and GPT-5.4.
M2.7 delivers outstanding real-world programming capabilities spanning log analysis, bug troubleshooting, refactoring, code security, and machine learning. Beyond code generation, M2.7 demonstrates strong system-level reasoning — correlating monitoring metrics, conducting trace analysis, verifying root causes in databases, and making SRE-level decisions. Using M2.7, we have reduced live production incident recovery time to under three minutes on multiple occasions.
On SWE-Pro, M2.7 achieved 56.22%, matching GPT-5.3-Codex, with even stronger performance on real-world engineering benchmarks: SWE Multilingual (76.5) and Multi SWE Bench (52.7). On VIBE-Pro (55.6%), M2.7 is nearly on par with Opus 4.6. On Terminal Bench 2 (57.0%) and NL2Repo (39.8%), M2.7 demonstrates deep understanding of complex engineering systems. M2.7 also supports native Agent Teams for multi-agent collaboration with stable role identity and autonomous decision-making.
M2.7 achieved an ELO score of 1495 on GDPval-AA (highest among open-weight models), surpassing GPT5.3. It handles Word, Excel, and PPT with high-fidelity multi-round editing, producing editable deliverables. On Toolathon, M2.7 reached 46.3% accuracy (global top tier), and maintains 97% skill compliance across 40+ complex skills on MM Claw. On the MM Claw end-to-end benchmark, M2.7 achieved 62.7%, close to Sonnet 4.6.
M2.7 features strengthened character consistency and emotional intelligence. We open-sourced OpenRoom, an interactive demo that places AI interaction within a Web GUI space with real-time visual feedback and scene interactions. Try it at openroom.ai.
Download the model from HuggingFace repository: https://huggingface.co/MiniMaxAI/MiniMax-M2.7
We recommend using the following inference frameworks (listed alphabetically) to serve the model:
We recommend using SGLang to serve MiniMax-M2.7. Please refer to our SGLang Deployment Guide.
We recommend using vLLM to serve MiniMax-M2.7. Please refer to our vLLM Deployment Guide.
We recommend using Transformers to serve MiniMax-M2.7. Please refer to our Transformers Deployment Guide.
You also can get model weights from modelscope.
MiniMax M2.7 is also available on NVIDIA NIM Endpoint.
We recommend using the following parameters for best performance: temperature=1.0, top_p = 0.95, top_k = 40. Default system prompt:
You are a helpful assistant. Your name is MiniMax-M2.7 and is built by MiniMax.
Please refer to our Tool Calling Guide.
Contact us at model@minimax.io.
model-00000-of-00130.safetensors 3.7 GBmodel-00020-of-00130.safetensors 2.5 GBmodel-00022-of-00130.safetensors 2.5 GBmodel-00024-of-00130.safetensors 2.5 GBmodel-00026-of-00130.safetensors 2.5 GBmodel-00028-of-00130.safetensors 2.5 GBmodel-00030-of-00130.safetensors 2.5 GBmodel-00032-of-00130.safetensors 2.5 GBmodel-00034-of-00130.safetensors 2.5 GBmodel-00036-of-00130.safetensors 2.5 GBmodel-00038-of-00130.safetensors 2.5 GBmodel-00040-of-00130.safetensors 2.5 GBmodel-00042-of-00130.safetensors 2.5 GBmodel-00044-of-00130.safetensors 2.5 GBmodel-00046-of-00130.safetensors 2.5 GBmodel-00048-of-00130.safetensors 2.5 GBmodel-00050-of-00130.safetensors 2.5 GBmodel-00052-of-00130.safetensors 2.5 GBmodel-00054-of-00130.safetensors 2.5 GBmodel-00056-of-00130.safetensors 2.5 GBmodel-00058-of-00130.safetensors 2.5 GBmodel-00060-of-00130.safetensors 2.5 GBmodel-00062-of-00130.safetensors 2.5 GBmodel-00064-of-00130.safetensors 2.5 GBmodel-00066-of-00130.safetensors 2.5 GBmodel-00068-of-00130.safetensors 2.5 GBmodel-00070-of-00130.safetensors 2.5 GBmodel-00072-of-00130.safetensors 2.5 GBmodel-00074-of-00130.safetensors 2.5 GBmodel-00076-of-00130.safetensors 2.5 GBmodel-00078-of-00130.safetensors 2.5 GBmodel-00080-of-00130.safetensors 2.5 GBmodel-00082-of-00130.safetensors 2.5 GBmodel-00084-of-00130.safetensors 2.5 GBmodel-00086-of-00130.safetensors 2.5 GBmodel-00088-of-00130.safetensors 2.5 GBmodel-00090-of-00130.safetensors 2.5 GBmodel-00092-of-00130.safetensors 2.5 GBmodel-00094-of-00130.safetensors 2.5 GBmodel-00096-of-00130.safetensors 2.5 GBmodel-00098-of-00130.safetensors 2.5 GBmodel-00100-of-00130.safetensors 2.5 GBmodel-00102-of-00130.safetensors 2.5 GBmodel-00104-of-00130.safetensors 2.5 GBmodel-00106-of-00130.safetensors 2.5 GBmodel-00108-of-00130.safetensors 2.5 GBmodel-00110-of-00130.safetensors 2.5 GBmodel-00112-of-00130.safetensors 2.5 GBmodel-00114-of-00130.safetensors 2.5 GBmodel-00116-of-00130.safetensors 2.5 GBmodel-00118-of-00130.safetensors 2.5 GBmodel-00120-of-00130.safetensors 2.5 GBmodel-00122-of-00130.safetensors 2.5 GBmodel-00002-of-00130.safetensors 2.5 GBmodel-00004-of-00130.safetensors 2.5 GBmodel-00006-of-00130.safetensors 2.5 GBmodel-00008-of-00130.safetensors 2.5 GBmodel-00010-of-00130.safetensors 2.5 GBmodel-00012-of-00130.safetensors 2.5 GBmodel-00014-of-00130.safetensors 2.5 GBmodel-00016-of-00130.safetensors 2.5 GBmodel-00018-of-00130.safetensors 2.5 GBmodel-00124-of-00130.safetensors 1.2 GBmodel-00021-of-00130.safetensors 1.2 GBmodel-00023-of-00130.safetensors 1.2 GBmodel-00025-of-00130.safetensors 1.2 GBmodel-00027-of-00130.safetensors 1.2 GBmodel-00029-of-00130.safetensors 1.2 GBmodel-00031-of-00130.safetensors 1.2 GBmodel-00033-of-00130.safetensors 1.2 GBmodel-00035-of-00130.safetensors 1.2 GBmodel-00037-of-00130.safetensors 1.2 GBmodel-00039-of-00130.safetensors 1.2 GBmodel-00041-of-00130.safetensors 1.2 GBmodel-00043-of-00130.safetensors 1.2 GBmodel-00045-of-00130.safetensors 1.2 GBmodel-00047-of-00130.safetensors 1.2 GBmodel-00049-of-00130.safetensors 1.2 GBmodel-00051-of-00130.safetensors 1.2 GBmodel-00053-of-00130.safetensors 1.2 GBmodel-00055-of-00130.safetensors 1.2 GBmodel-00057-of-00130.safetensors 1.2 GBmodel-00059-of-00130.safetensors 1.2 GBmodel-00061-of-00130.safetensors 1.2 GBmodel-00063-of-00130.safetensors 1.2 GBmodel-00065-of-00130.safetensors 1.2 GBmodel-00067-of-00130.safetensors 1.2 GBmodel-00069-of-00130.safetensors 1.2 GBmodel-00071-of-00130.safetensors 1.2 GBmodel-00073-of-00130.safetensors 1.2 GBmodel-00075-of-00130.safetensors 1.2 GBmodel-00077-of-00130.safetensors 1.2 GBmodel-00079-of-00130.safetensors 1.2 GBmodel-00081-of-00130.safetensors 1.2 GBmodel-00083-of-00130.safetensors 1.2 GBmodel-00085-of-00130.safetensors 1.2 GBmodel-00087-of-00130.safetensors 1.2 GBmodel-00089-of-00130.safetensors 1.2 GBmodel-00091-of-00130.safetensors 1.2 GBmodel-00093-of-00130.safetensors 1.2 GBmodel-00095-of-00130.safetensors 1.2 GBmodel-00097-of-00130.safetensors 1.2 GBmodel-00099-of-00130.safetensors 1.2 GBmodel-00101-of-00130.safetensors 1.2 GBmodel-00103-of-00130.safetensors 1.2 GBmodel-00105-of-00130.safetensors 1.2 GBmodel-00107-of-00130.safetensors 1.2 GBmodel-00109-of-00130.safetensors 1.2 GBmodel-00111-of-00130.safetensors 1.2 GBmodel-00113-of-00130.safetensors 1.2 GBmodel-00115-of-00130.safetensors 1.2 GBmodel-00117-of-00130.safetensors 1.2 GBmodel-00119-of-00130.safetensors 1.2 GBmodel-00121-of-00130.safetensors 1.2 GBmodel-00123-of-00130.safetensors 1.2 GBmodel-00001-of-00130.safetensors 1.2 GBmodel-00003-of-00130.safetensors 1.2 GBmodel-00005-of-00130.safetensors 1.2 GBmodel-00007-of-00130.safetensors 1.2 GBmodel-00009-of-00130.safetensors 1.2 GBmodel-00011-of-00130.safetensors 1.2 GBmodel-00013-of-00130.safetensors 1.2 GBmodel-00015-of-00130.safetensors 1.2 GBmodel-00017-of-00130.safetensors 1.2 GBmodel-00019-of-00130.safetensors 1.2 GB有文件？提交种子。
提交种子
LocalMaxxing测试者没有提供启动命令。
在 LocalMaxxing 上查看测试docker run --rm -d --name lsv-mm3-u-tp4 --net=host --shm-size=32g --device /dev/dri:/dev/dri -v /dev/dri:/dev/dri -v /models/weights:/llm/models:ro -v /models/bench/lmx-models-s6b/arms/mm3-u-tp4/session:/bench -e ZE_AFFINITY_MASK=0\,1\,2\,3 -e VLLM_XPU_ENABLE_XPU_GRAPH=1 -e DISABLE_ESIMD_GDN_OUTPROJ=1 -e DISABLE_ESIMD_INT4=0 -e DISABLE_ESIMD_INT4_GEMM=0 -e VLLM_OFFLOAD_WEIGHTS_BEFORE_QUANT=1 -e VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 -e VLLM_WORKER_MULTIPROC_METHOD=spawn -e CCL_ZE_IPC_EXCHANGE=sockets --entrypoint vllm vllm/vllm-openai-xpu:nightly-77154ca6116ae34cbfe314beef1016bde9b63989@sha256:a410323607625d551d370253e4c5e1a038bbdf8ba2535c2a30bdf5ac531fc4f4 serve --model /llm/models/MiniMax-M2.7-REAP-172B-W4A16-gptq --host 0.0.0.0 --served-model-name minimax-m27 --tensor-parallel-size 4 --pipeline-parallel-size 1 --data-parallel-size 1 --gpu-memory-utilization 0.92 --max-model-len 4096 --max-num-seqs 4 --max-num-batched-tokens 8192 --block-size 64 --kv-cache-dtype auto --mamba-ssm-cache-dtype auto --dtype float16 --quantization gptq --no-enable-prefix-caching --trust-remote-code --cpu-offload-gb 0 --port 8000在 LocalMaxxing 上查看测试
VLLM_XPU_USE_LLM_SCALER_MOE=1 CCL_TOPO_P2P_ACCESS=1 VLLM_XPU_ENABLE_XPU_GRAPH=0 VLLM_CACHE_ROOT=/mnt/fast-ai/vllm-cache-exp/minimax-inductor-partition-compile1-p512n512-20260512T122329Z vllm bench throughput --backend vllm --async-engine --model /mnt/fast-ai/llm-models/minimax-m2.7-int4-autoround --tokenizer /mnt/fast-ai/llm-models/minimax-m2.7-int4-autoround --trust-remote-code --dtype float16 --tensor-parallel-size 4 --distributed-executor-backend mp --max-model-len 2048 --max-num-batched-tokens 1024 --max-num-seqs 1 --dataset-name random --random-input-len 512 --random-output-len 1536 --random-range-ratio 0 --num-prompts 1 --disable-log-stats --compilation-config '{"use_inductor_graph_partition":true,"compile_sizes":[1]}'在 LocalMaxxing 上查看测试 测试者没有提供启动命令。
在 LocalMaxxing 上查看测试/home/mikekey/Experiments/llama.cpp/build-vulkan/bin/llama-server --port ${PORT} --host 0.0.0.0 --no-webui -ngl 99 --jinja --cache-reuse 256 --parallel 2 --prio 3 -m /home/mikekey/models/gguf/minimax/M2.7-REAP-172B-IQ4_XS/MiniMax-M2.7-REAP-172B-A10B-BF16.i1-IQ4_XS.gguf --temp 1.0 --top-p 0.95 -ctk q8_0 -ctv q8_0 --flash-attn auto -ts 1/1/1 --main-gpu 1 -c 16384在 LocalMaxxing 上查看测试 vllm serve /root/minimax-m27-weights -tp 2 --distributed-executor-backend ray --gpu-memory-utilization 0.85 --max-model-len 190000 --max-num-seqs 4 --max-num-batched-tokens 8192 --enable-prefix-caching --enable-chunked-prefill --kv-cache-dtype fp8 --attention-backend flashinfer --dtype auto --disable-custom-all-reduce --load-format fastsafetensors --tool-call-parser minimax_m2 --reasoning-parser minimax_m2_append_think --trust-remote-code在 LocalMaxxing 上查看测试
来自 LocalMaxxing 的 112 次社区测试（7 种配置）的中位数。
自己运行：配置指南