Daily AI Launch Radar: Mistral Large 4 weights, Copilot Fusion and Replit Cloud Agents

Three updates in this edition give developers new choices for self-hosting a model, routing coding requests and proposing changes to a company codebase. The launch dates run from October 27 to October 29; Kingy reviewed the sources but did not test these products.

Village Global photograph titled “Amjad Masad & Adam D’Angelo,” taken at The Grove on November 17, 2022. Historical Replit company context for a Radar covering Replit Cloud Agents alongside Mistral and GitHub. This photograph does not show any of those products or their launches. Photo: Village Global / Wikimedia Commons, CC-BY-2.0. Cropped to 16:9 and resized to 1600×900; photographic content otherwise unchanged. No photographer, participant or vendor endorsement implied. Only the photograph is licensed under CC-BY-2.0; the license does not cover this article.

Mistral Large 4: downloadable weights with conditions

Mistral released improved Large 4 (Le Chonk) weights on October 29, following an earlier public preview. The model card describes text and image input, text output and a one-million-token context. Teams can compare the hosted API with self-hosting, but the downloadable weights carry a restricted license: a licensee whose company or affiliate operates a Model-as-a-Service business must obtain a separate commercial license before commercial use, with exceptions for internal use and official Mistral products.

Mistral lists API prices of US$0.70 input, US$0.07 cached input and US$2.10 output per million tokens. Self-hosting compute is separate. Kingy did not run the model or verify hardware requirements. Read the Kingy launch record and Mistral’s announcement.

GitHub Copilot Fusion: model routing in supported clients

GitHub made Copilot Fusion generally available on October 28 across supported Copilot clients. It can route a request to one model, escalate a draft or request critique; available models and effort settings depend on plan and administrator policy.

Underlying models are billed at their standard rates, with no separate Fusion charge, according to the current Copilot Fusion documentation. GitHub documents a 30% paid-plan model-cost discount through November 30, 2026; multi-model tasks can use more credits. Kingy has not tested routing quality, latency or costs. Review generated changes before committing. Read the Kingy launch record and GitHub’s announcement.

Replit Cloud Agents: proposed changes for engineering review

Replit opened its company-codebase Cloud Agent workflow to Pro and Enterprise customers on October 27. Contributors can request changes in plain language, with proposed code delivered as GitHub pull requests or GitLab merge requests for engineers to review.

Paid-plan availability is documented, but the reviewed pricing page did not establish plan amounts or a Cloud-Agent-specific usage rate. Check current account terms and review generated code; Kingy did not test the workflow or deployment safety. Read the Kingy launch record and Replit’s announcement.