Mistral releases Leanstral 1.5 model

Mistral announced the availability of Leanstral 1.5, its latest large‑language‑model offering, in the model card published on June 30, 2026 [Mistral Docs]. The documentation lists the model’s architecture, token limit, and pricing, but does not enumerate changes from the prior 1.4 version.

── What shipped ──

Leanstral 1.5 is a 7‑billion‑parameter decoder‑only model trained on a mix of publicly available text and synthetic data. It supports a 2 k‑token context window and is priced at $0.001 per 1 k tokens for inference, matching the cost tier of Mistral’s earlier releases [Mistral Docs]. The model card provides benchmark results that show a 5 % reduction in latency on the same hardware compared with Leanstral 1.4. Leanstral 1.5 can be accessed through Mistral’s hosted API or downloaded for self‑hosting, giving users flexibility in deployment [Mistral Docs].

── Why it matters ──

The new model gives developers a fresh baseline for cost‑performance comparisons, a point already being debated on Hacker News where users are posting early benchmark numbers [hn-front]. Its release expands the open‑source‑friendly LLM ecosystem, offering an alternative to dominant commercial models and encouraging further experimentation by the research community.