RWT
9.5 score
LLM Leaderboard · Open Weight
MiniMax-M3 benchmarks, pricing, and LLM comparison.
Compare MiniMax-M3 vs GPT, Claude, Gemini, DeepSeek, open-weight, and frontier AI models using public benchmark scores, token pricing, context window, and access details.
Rank #9AskClash overall score: 67.4
$0.30 / $1.20Input and output token price, when published. Context: 1M.
API/OAuthBilling and access path cached for this model row.
MiniMax-M3 benchmark snapshot
AskClash combines public LLM benchmark cells into a weighted percentile score and penalizes missing coverage so narrow rows do not dominate better-measured models.
Overall67.4
Benchmark cells13
Context1M
CreatorMiniMax
MiniMax-M3 public benchmark scores
Cached benchmark values can include HLE, GPQA, SWE-bench, SWE-Pro, SWE-Atlas, Terminal-Bench, MCP Atlas, MMMU-Pro, ARC-AGI-2, Tau2, and model-specific coding or agent scores.
HLE
37.1 score
GPQA
92.9 score
IFEval
82.9 score
SWE-bench
80.5 score
SWE-Pro
59.0 score
Terminal-Bench
66.0 score
OSWorld
70.1 score
MCP Atlas
74.2 score
Finance Agent
48.3 score
MMMU-Pro
78.1 score
Tau2
88.9 score
MiniMax-M3 vs other AI models
Use these comparison links to evaluate MiniMax-M3 against nearby LLMs by benchmark score, price, context window, and provider.
Related AI and tech coverage
Cached AskClash article matches that can provide release, provider, benchmark, pricing, or market context around this model.
HuggingFace Transformers v5.12.0 Release Notes
MiniMax-M3-VL is the vision-language member of the MiniMax-M3 family that pairs a CLIP-style vision tower with 3D rotary position embeddings with the MiniMax-M3 text backbone. It uses a mixed dense/sparse Mixture-of-Experts decoder with SwiGLU-OAI gated experts and a lightning indexer for block-sparse attention. The model processes images through a Conv3d patch embedding system and includes specialized components for efficient multimodal understanding and generation. The official weights for PP-
vLLM Inference Engine v0.19.0 Release Notes
* Model fixes: NemotronH MTP + Chunked Prefill (#35447), Qwen3-VL video timestamps (#37439), Qwen3.5 GDN quantized models (#37448), Qwen3Next A_log FP32 (#37810), JAIS ALiBi (#37820), RoBERTa CUDA graph position IDs (#37873), AudioFlamingo3/MusicFlamingo (#37643), Music Flamingo loading (#35535), bge-m3 task selection (#37632), Nemotron Parse loading (#37407), GLM OCR patch merger (#37962), PaddleOCR checkpoint compat (#38232), DeepSeek v3.2 params (#33703), MiniMax NVFP4 weight loading (#37214)
vLLM Inference Engine v0.23.0 Release Notes
* **DeepSeek-V4 matures across backends**: Following its introduction in v0.22.0, DeepSeek-V4 received another large hardening and optimization pass. Its sparse MLA metadata is now decoupled from DeepSeek-V3.2 (#44699), it gained a TRTLLM-gen attention kernel (#43827), EPLB support for the Mega-MoE (#43339), selective prefix-cache retention for sliding-window KV cache (#43447), and an index-share feature for DSA MTP (#44420). The model was also detached from `torch.compile` (#43746, #43891), its
Last cached leaderboard date: June 1, 2026. This model page is generated from the AskClash LLM Leaderboard cache and linked from the live leaderboard.