US Lifts Export Controls on Anthropic; Sonnet 5 Debuts with 1M‑Token Context but Higher Costs — OpenAI Cuts ChatGPT Fees, Google Clears Gemini 3.5 Pro

Share your love

AI Tech News Today — Aurora & Isabelle

Timestamp: 2026-07-01T10:07:40.000Z
Top A‑I stories you need to know — about two minutes.


1) United States lifts export controls for Anthropic models

What happened

  • The U.S. Department of Commerce eased export restrictions, allowing Anthropic’s flagship systems to come back online.
  • Claude Fable 5 and Mythos 5 may return after a multi‑week ban tied to jailbreak concerns.

Why it matters

  • Fable 5 is expected to be available later today, though it’s unclear whether access will extend beyond the API.
  • Anthropic may be required to implement stricter identity checks to satisfy regulators.

2) Anthropic launches Sonnet 5 — big context window, surprising costs

Overview

  • Sonnet 5 is now Anthropic’s default mid‑tier model.
  • Key features: 1,000,000 token context window, real‑time cyber safeguards, and five reasoning effort settings for complex coding and agent workflows.

Notable twist

  • Early benchmarks show Sonnet 5 can cost more per completed task than Anthropic’s premium model Opus 4.8, even with promotional pricing:
    • $2 per million input tokens (promo)
    • $10 per million output tokens (promo) through August

Other Anthropic updates

  • Claude Science in beta for macOS and Linux.
  • A self‑hosted Claude Apps Gateway for enterprise deployments.

3) OpenAI teases developer hardware for Codex and cuts ChatGPT costs

Cost and infrastructure

  • OpenAI quietly reduced inference costs for logged‑out ChatGPT traffic by more than half.
  • This freed thousands of GPUs and cut the active GPU count for guest users to a few hundred.

Engineering & benchmarks

  • Engineers fixed long‑standing crashes in search pipelines.
  • Launched GeneBench‑Pro, a new benchmark for computational biology tasks.

Hardware tease

  • OpenAI teased a compact developer macro pad (built with Work Louder) aimed at speeding Codex coding workflows — dropping July 15.

4) Google clears Gemini 3.5 Pro and expands creative tools

Model releases

  • Gemini 3.5 Pro greenlit for a July release on Google Cloud.
  • Google is adding specialized quantitative models to its marketplace.

Creative and productivity tools

  • Gemini Spark evolving into a macOS desktop agent to automate files and app workflows.
  • New media models:
    • Nano Banana 2 Lite — lightning‑fast image model.
    • Gemini Omni Flash — supports iterative, chat‑style video editing at ~$0.10/second.
  • Notebook L‑M is getting mobile features that convert dense research into 60‑second explainers.

That’s the roundup for now.
Stay curious, stay skeptical — join us tomorrow for more A‑I news.
This is Aurora and Isabelle — thanks for listening.

Share your love