OpenAI Previews GPT-5.6 While Anthropic Restores Fable Five and Rolls Out Claude Sonnet 5; Meta Debuts Brain-to-Qwerty v2 as Etched Ships Custom Inference Racks

AI Tech News Today — Quick Roundup
Date: 2026-07-03T11:30:53.000Z
Hosts: Aurora and Isabelle
Runtime: About 2–3 minutes

1. Claude Fable Five — access restored

   • Anthropic says it has restored access to Fable Five after conversations with the U.S. government and additional cybersecurity hardening.
   • Paid plans receive included usage through July 7 at up to 50% of weekly allotment; after that, Fable Five will be available only via credits.
   • Takeaway: a cautious reopening and a reminder that national security talks are shaping which models people can use and how.

2. OpenAI previews GPT‑5.6

   • Next‑gen models previewed, led by a powerful Sol variant built for agent‑style tasks in coding, biology, and cybersecurity.
   • New features:
     • Max reasoning effort setting
     • Ultra mode that spins up subagents for complex workflows
     • Tiered lineup: Sol, Terra, and Luna at different price points
   • Preview limited to trusted partners for now; wider rollout expected soon.

3. Meta advances brain‑to‑text

   • Meta released Brain to Qwerty v2, a noninvasive system that decodes full sentences and semantic meaning in real time from raw magnetoencephalography (MEG) signals.
   • Trained on thousands of volunteer sentences, it improves with more data and shows strong accuracy.
   • Meta is open‑sourcing the training code to accelerate neuroscience research — which could speed development but also raises fresh ethical questions.

4. Anthropic launches Claude Sonnet Five

   • Claude Sonnet Five is announced as Anthropic’s strongest Sonnet model yet.
   • Improvements: reasoning, tool use, coding, and knowledge work — narrowing the gap with Opus 4.8 at a lower price.
   • It’s now the default on Free and Pro plans across all Claude surfaces, so users see the upgrade immediately.

5. Etched exits stealth with custom inference hardware

   • Startup Etched unveiled production racks of custom AI inference hardware after a successful chip tapeout.
   • Company says it has >$1 billion in signed contracts and $800 million raised.
   • Design focus: low‑voltage inference and cluster‑scale memory to improve throughput on large models; first racks ship this summer.

Stay informed

• That’s it for today — concise, current, and a little cheeky. Keep listening and reading to follow how AI is changing software, hardware, and regulation.
  From Aurora and Isabelle, thanks for joining AI Tech News Today.