Revision History
Mistral Large 3 · 13 revisions
Sizes are character counts of the article source. The signed number is the change from the previous revision.
Recent edit summaries
Detailed summaries recorded by editors. Generic maintenance summaries are omitted here; every recorded revision remains below. Dates describe the edit, not necessarily the event it covers.
- Correction: Corrected stale claims that a reasoning Mistral Large 3 was still coming (it never shipped and Mistral Large 4 succeeded it in October 2026), the tokenizer (Tekken, not a non-Tekken tokenizer), Artificial Analysis figures that reversed its above-average ratings and misstated the token median, an unsupported MMLU-Pro score, a GPT-5 context comparison that contradicted the article's own table, the Medium 3.1 release date, and a 'tripled revenue' figure misattributed to Large 3 that was actually from May 2025 (plus 35 minor fixes)
Version 13 · Nov 4, 2026, 08:13 PM
- Correction: dated the competing MoE flagships (DeepSeek V3 December 2024, Kimi K2.5 January 2026, DeepSeek V4 April 2026) instead of placing all of them in 2025, and updated the DeepSeek V4 comparison row to V4-Pro's published 1.6T total / 49B active parameters and 1M-token context.
Version 12 · Oct 31, 2026, 04:03 AM
- Correction: the released checkpoint's params.json gives the expert count and routing (128 routed experts plus one shared, four per token) and shows Multi-head Latent Attention.
Version 11 · Oct 31, 2026, 03:47 AM
- Correction: replaced the stale expectation about a future Large 4 with the dated Mistral Large 4 preview (October 6, 2026) and open-weight release (October 29, 2026).
Version 10 · Oct 30, 2026, 10:49 AM
- Start the article with its lead paragraph: moved the first paragraph above the "What is Mistral Large 3?" heading. No text changed.
Version 9 · Oct 29, 2026, 09:23 AM