Unverified50% confidenceFactExact time
Compared with MiniMax M2.7, M3.1 Flash's context window expands from 204,800 to 1,000,000, its input types expand from text-only to text, images, and video, and its thinking mode changes from always-on with no levels to five adjustable levels defaulting to max.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/19/2026
First Seen
Valid until: 1/17/2027
Sources
Related Entities
Related Claims
UnverifiedMiniMax M3.1 Flash defaults to the max setting when the effort field is not passed in the request, and the model always performs thinking.70% similarUnverifiedMiniMax M3.1 Flash's thinking tokens count toward the max_tokens limit, so reusing the 4K or 8K caps from the M2.x era may cause truncation with finish_reason: "length"; MiniMax recommends raising max_tokens to 131,072, with an upper limit of 524,288.68% similarUnverifiedDifferent embedding models have different output dimensions: paraphrase-multilingual-MiniLM-L12-v2 is 384 dimensions, the BERT-base series is 768 dimensions, OpenAI text-embedding-3-small is 1536 dimensions, and text-embedding-3-large and Google gemini-embedding-001 are 3072 dimensions.68% similarUnverifiedStep 3.5 Flash支持256,000个token的上下文窗口67% similarUnverifiedMaximum output tokens: MiniMax M3.1 Flash is 512K, DeepSeek V4.1 Flash 1017 is 384K, and Claude Haiku 5.5 is 128K.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067219API
curl https://kongchang.com/api/v1/knowledge/claims/1067219MCP
get_claim(id=1067219)