Evening brief

Thursday, July 2

3 people · 0 companies · 5 topics

TL;DR Anthropic had the biggest 48 hours of its year: Fable 5 + Mythos 5 came back globally (export controls lifted June 30, access restored July 1), and Claude Sonnet 5 shipped as the new default across Claude Code and all user tiers. Meanwhile GPT-5.6 Sol is still gated to ~20 partner orgs with no GA date, and Gemini 3.5 Pro keeps slipping.

Models & research

  • model-releases Fable 5 and Mythos 5 are back globally as of July 1, ending a 19-day suspension triggered by US export controls applied on June 12. Pro/Max/Team users get up to 50% of weekly usage limits free through July 7, then usage credits. Anthropic · Al Jazeera · CNBC

  • model-releases Claude Sonnet 5 (released June 30) is now the default model for all Claude users and the new default in claude-code, with a 1M-token context window and introductory pricing of $2/$10 per Mtok through August 31. Described as the most agentic Sonnet yet, it runs close to Opus 4.8 on many tasks at a fraction of the cost. AI Business

  • model-releases GPT-5.6 Sol/Terra/Luna preview (announced June 26) is still locked to roughly 20 government-cleared orgs. Sol ($5/$30 per Mtok) targets hard reasoning and agentic coding; Terra ($2.50/$15) is the everyday workhorse; Luna ($1/$6) is lowest-cost. OpenAI says Sol will launch on Cerebras at up to 750 tokens/sec in July, with broad GA expected “in coming weeks.” OpenAI · VentureBeat

  • model-releases Gemini 3.5 Pro slipped from June to July GA. Standout specs: 2M-token context, Deep Think reasoning (Ultra tier only). The delay coincides with four senior Google researchers departing for OpenAI and Anthropic, including Gemini co-lead Noam Shazeer. TechTimes · Bind AI

Tools & dev

  • claude-code Latest Claude Code update ships session reliability overhaul: long-running agent sessions now survive process restarts on all platforms; stalled agents get labeled “Needs attention”; completed rows no longer flip state. Also adds org default models, clickable file attachments, and a 25% token reduction in the /code-review workflow. Releasebot

  • agentic-ai emollick cited new OpenAI internal data showing the chatbot era is functionally over inside the company: agentic task completion is spreading beyond engineering to knowledge work broadly. “The chatbot era is over, and agentic systems are coming to tasks beyond engineering.” X

  • dev-tools The CREATE AI Act, a rare bipartisan bill, is moving through Congress. It aims to establish federal standards for AI deployment across government and industry. Fortune

  • steipete “Loop engineering” framing — Peter Steinberger’s June 8 post (“you shouldn’t be prompting coding agents anymore, you should be designing loops that prompt your agents”) hit 6.5M views on X and is still circulating widely among agentic dev communities. ExplainX

OSS

  • oss OmniRoute (diegosouzapw/OmniRoute) is trending today (+1,010 stars): a free AI gateway that routes 231+ providers through a single OpenAI-compatible endpoint, with RTK+Caveman token compression saving 15-95% and native support for Claude Code, Codex, Cursor, Cline, and Copilot. GitHub

  • oss agency-agents (msitarzewski) is the day’s biggest mover (+2,114 stars): a self-described “complete AI agency” toolkit covering frontend, community management, and content agents out of the box. GitHub

  • oss strix (usestrix) trending (+1,211 stars): open-source AI penetration testing tool for finding and fixing app vulnerabilities. GitHub