Bulletin · UTC
Merged timeline: 10 items (blog publish time and listing createdAt in UTC). For registry-only weekly slices, use /new.
- Skillhumanizer
Remove signs of AI-generated writing to enhance text's natural flow.
- Skillimprove
Survey codebases and produce implementation plans for improvements.
Cohere's first agentic coding model designed for developers. It combines efficiency with powerful coding capabilities, making it ideal for modern software engineering tasks.
Claude Fable 5 is a next-generation intelligence model designed for ambitious work. It excels at long-running tasks and can investigate codebases before acting.
Launched June 9, 2026, Claude Fable 5 demonstrates stunning 3D worldbuilding capabilities, creating Minecraft clones with multiple biomes, caves, and ore systems in 20-55 minutes from a single prompt—all using custom browser-based ThreeJS implementations.
Claude Fable 5 and Mythos 5 are live, offering advanced agentic autonomy, SOTA coding, vision, and genomics capabilities starting at $10 per million input tokens.
Launched June 9, 2026, North Mini Code is Cohere's first open-source agentic coding model—a 30B parameter mixture-of-experts model with just 3B active parameters. Available under Apache 2.0, it delivers competitive performance on SWE-Bench and Terminal-Bench 2.0 while offering 2.8x higher output throughput than Devstral Small 2.
Lance Martin from Anthropic shares insights on designing loops with Claude Fable 5: self-correction loops with /goal and Outcomes primitives, verifier sub-agents that outperform self-critique, memory management across sessions, and rubric design principles that achieve 6x improvements on Parameter Golf over Opus 4.7.
Claude Fable 5 is fully available on Claude Code, bringing state-of-the-art intelligence to your coding workflows. Run multi-day projects, implement complex migrations, and leverage agent harnesses for autonomous work.
Published June 8, 2026, Self-Harness demonstrates how AI agents can autonomously identify weaknesses, propose harness modifications, and validate improvements—turning model-specific failure patterns into concrete executable fixes that boost Terminal-Bench 2.0 pass rates from 40.5% to 61.9%, 23.8% to 38.1%, and 42.9% to 57.1% across three diverse models.