ModelsMiniMax M1: An Open Reasoning Model Built for Long Outputs
MiniMax M1 is a 456B open-weight reasoning model with a 1M window and hybrid attention that keeps long thinking cheap. What it claims, what it costs and how to run it.
ReadNotes on models, agents, and what inference costs
Practical writing for developers building with large language models — how they work, how to pick one, and how to keep the bill predictable.
Models
ModelsMiniMax M2 Guide: The 10B-Active Model That Started the Line
MiniMax M2 put a 230B MoE with about 10B active behind a cheap API in October 2025. What it is, the thinking-history trap, and where it still earns a place.
Read
ModelsMiniMax M2-her: A Dialogue Model for Roleplay, Not Coding
MiniMax M2-her is built for character chat, not agents: a 64K window, a 2,048-token reply cap and extra roles for personas and example dialogue. How to use it well.
Read
ModelsMiniMax M2.1: What a Point Release Changed for Coding Agents
MiniMax M2.1 kept M2's 10B-active design and retrained it for coding beyond Python, mobile apps and tool use. The gains, the interleaved thinking trap, and the cost.
ReadModelsMiniMax M2.5: A Strong Coding Model With a Retirement Date
MiniMax M2.5 reported 80.2% on SWE-bench Verified in February 2026. MiniMax now lists it as legacy and retires it on its own API in November. What that means.
ReadModelsMiniMax-01: The Lightning-Attention Model Behind the Family
MiniMax-01 pairs a 456B hybrid-attention text model with a vision encoder and a 1M window. How lightning attention works, what it costs and when it still makes sense.
ReadModelsMinistral 3 14B Guide: A Local Model That Sees and Calls Tools
Ministral 3 14B is an Apache 2.0 dense model with vision, a 256K window and FP8 weights that fit in about 24GB. When it beats Mistral Small, and when to drop to the 8B.
ReadModelsMinistral 3 3B: A Vision Model Small Enough for the Edge
Mistral's smallest Ministral 3 model reads images, calls tools and fits in 8GB of VRAM. What it is good for, where 3B runs out, and when to step up to 8B.
Read
ModelsMinistral 3 8B: The Small Mistral Model to Start With
Ministral 3 8B reads images, calls tools, runs in about 12GB of VRAM and costs the same per token for input and output. Where it fits between the 3B and 14B.
ReadModelsMistral Large 2 vs Large 3: Very Different Models, One Name
Mistral Large 2 was a 123B dense model under a research licence. Mistral Large 3 is a 675B MoE under Apache 2.0 at a quarter of the price. Know which you are calling.
ReadModelsMistral Medium 3: What It Was, and Planning Around Its Retirement
Mistral Medium 3 launched in May 2025 as a mid-priced enterprise model and is now past its retirement date. What it offered, and how to move off it cleanly.
Read
ModelsMistral Medium 3.1: Evaluating a Point Release With Few Numbers
Medium 3.1 kept Medium 3's price and window and promised better tone and performance, with little published detail. How to judge a release like that, and what replaces it.
ReadShowing 193–204 of 731 articles