OpenJDK bans AI-generated code; Oracle's GraalVM allows it
Effective Aug 4, the JDK rejects any LLM-written code—even partly edited. Plus: Copilot review tiers, a 7-cent long-context model, and free GPT-5.6 Luna.
Copy markdown
No AI code in the JDK—not even one line
OpenJDK's interim policy, live since Aug 4, bars any contribution “created in whole or in part” by LLMs or diffusion models—code, docs, PRs, emails, even bug reports. Editing 10 of 100 AI-written lines still counts as banned. Contributors must tick a compliance box in the Skara review system.
GraalVM, same owner, opposite call
Oracle's other JVM project welcomes AI tools for drafting, transforming, and reviewing code—as long as you understand and can defend the change. Disclosure is encouraged, not required; if you can't explain an AI-assisted change, it gets rejected.
Copilot review depth is now a dial
GitHub made code-review effort levels GA: pick “Lite” for docs and small fixes, “Balanced” for risky cross-service changes. It's per-PR, admins set org defaults, and the level is stamped on the review. Old Low/Medium settings auto-migrated.
A 262K-context model for 7 cents
inclusionAI's Ling 3.0 Flash is live on Vercel AI Gateway and DeepInfra at $0.07 in / $0.22 out per million tokens with 262K context; the Tiny variant is free on the Gateway through mid-August. Cheap long-context for RAG pipelines and agents.
GPT-5.6 Luna goes free
OpenAI expanded free-tier access to GPT-5.6 Luna—reportedly unlimited text chats—and added a reasoning-effort slider to Sol, so free users get a stronger default and paid users can trade speed for depth per prompt.