Kimi
Moonshot AI
Follow the jump from a coding update to a new attention stack.
Mistral Large 4 / Lab profile
Mistral AI
Choose the artifact before choosing the GPU count.
First time comparing models? Learn the terms before the architecture →
Mistral Large 4 is the Mistral checkpoint covered in this edition: a 1T-parameter multimodal MoE released as open weights at the end of October 2026 after an October 6 public preview. Mistral publishes it under its Ecosystem Open-Weight License, a modified MIT license.
Architecture trajectory
What connects these models?
The text model routes each token to four of 128 experts plus one shared expert, uses latent-attention projections and adds a sparse-attention indexer, with a Pixtral-style vision encoder for image input. Mistral reports 49B active parameters, or 52B counting embeddings and the output layer.
Store the history. Read the selected parts.
Fewer bytes read can reduce bandwidth pressure.
That does not prove an equal reduction in stored bytes.
Newest first / grouped sequence, not a time scale
The Mistral Large 4 timeline
Mistral-Large-4-1T-A52B
A European 1T open-weight MoE with two artifact paths
It combines 128 routed experts with four active and one shared expert, latent-attention projections and a sparse-attention indexer, at a 1M context setting. Mistral moved from an October 6 public preview to open weights at the end of the month.
Public preview announced 6 October with weights promised by month end; the public repository's first release commit is dated 29 October.
Which other labs should you compare?
Compare it with Kimi K3.1 Pro for large-cluster agent work and with GLM-5.3 for coding, under the same harness and reasoning budget. No cross-lab quality ranking is asserted here.
GLM
Z.ai
Track the attention changes separately from the post-training gains.
Interview preparation
What you should be able to explain
Work out weights per GPU for the FP8 and NVFP4 artifacts on eight GPUs, then explain why the H200 layout needs sixteen GPUs while a GB300 tray of four can run NVFP4.
Study the underlying systems → · Get the illustrated report →
Sources and scope
Checked 13 September 2026. Assessments and comparison candidates are our interpretation of the cited designs, not measured cross-lab rankings. Each timeline entry links to its release evidence; model pages preserve pinned card and configuration sources.