Mistral’s next few model generations could be much more important than people realize. Large 4 may only be the first step.
1
19
Mistral has been scaling European compute aggressively, and Arthur Mensch says future model generations could have access to around 20× more training compute. That doesn’t mean Large 4 itself got the full 20× boost.
1
27
My bet: Large 4 = major jump in architecture, post-training and agents Large 5 / 6 = where the much bigger compute base really kicks in If Mistral executes, the next 12 months could completely change how people view them.
7
Update: Mistral’s upcoming model now appears to be Large 4. Arthur Mensch says it’s coming “in days,” a notable shift from “in the coming weeks” just a few days ago.
9
Arthur Mensch says Mistral’s new large model is coming “in the coming weeks” and that it should close the gap with US labs “very significantly.”
1
9
Finally, an update on Mistral’s next frontier model. Arthur Mensch now says a new model is coming “in the coming weeks.” After missing the original summer timeline, this is the clearest signal yet that the release is getting close.
1
17
Meanwhile, Mistral has been upgrading everything around it: - Vibe - Agentic tooling - Regional inference - European compute - Third-party open models like GLM-5.3 The stack is increasingly ready for a much stronger model.
1
19
And that sets a very high bar. GLM-5.3 is already the default in Vibe and one of the strongest open models available. If Mistral’s own model replaces it as the default, it needs to be genuinely competitive. October could be very interesting.
10
GLM-5.3 becoming the default in Mistral Vibe sets a seriously high bar for Mistral’s upcoming frontier model. Their own model now has to beat one of the strongest open models in the world to take back the default spot.
23
Jack retweeted
GLM 5.3, hosted on Mistral, is now in Vibe Code, and the default in the web app. 1M context window and generous limits included. Happy 🛳️!
113
76
1,287
372,397
The summer timeline for Mistral’s next big open-weight model looks delayed. Arthur Mensch said early access would start in July, with the model coming “this summer.” Now mid-September, still no model, benchmarks, weights or release date. Still curious what they’re cooking.
1
33
What I expect Mistral to launch soon: A large, sparse, open-weight model targeting the US frontier. My guess: top-tier open-weight performance, around GLM-5.3 class or better, focused on reasoning, coding and agentic workloads. Could be Mistral’s biggest model jump yet.
1
1
39
Arthur Mensch said the upcoming model would start a new open-weight family, “fat indeed, but sparse.” Mistral’s CTO recently confirmed it was still training, while Mistral has been expanding compute, Vibe and its agentic stack. This doesn’t look like a minor update.
1
21
CLI v2.0.0 I've just published v2.0.0 of the ore-cli. You can upgrade today by running the following command: cargo install ore-cli 🧵
43
29
232
65,552
The v1.17.31 release is now recommended for general use by MainnetBeta validators. This release contains enhancements which will help alleviate the ongoing congestion on the Solana Network.
281
481
1,598
391,468
See the noticeable difference in fingerprints on the midnight color of the new M3 (left) vs the M2 (right) MacBook Air — all thanks to the improved anodization process! #MacBookAir
1
779
The Solana network is back up, there was a brief fork resolution period after supermajority was achieved but we are now seeing new rooted blocks and finalized slots advancing
26
49
322
49,220
Thrilled to announce that @circle will be participating in Jupiter Start for their new EURC token launch! They are participating in our community intro, where they will be and explaining their fresh EURC in our Discord (hint alpha coming) Also, $EURC is pre-listed on Jupiter, allowing you to prepare your orders now or trade it immediately afterward. We are super excited to help $EURC become a major stable in Solana ecosystem, complementing the anchor USDC that's already present everywhere!
97
269
1,593
108,807