In partnership with
Welcome back to The AI Field.
Anthropic had a huge day: a cheaper, near-Opus Claude that's now the free default, and its banned Fable 5 model is finally coming back online. Meanwhile, Meta showed off an AI that types words straight from your brain.
In today's AI Brief:
💬 Claude Sonnet 5 is the new free default
🔓 Fable 5 returns as the US lifts its ban
🧠 Meta's AI types what you're thinking
Read time: 4 minutes.
Books closed Sunday. You found out Monday morning.
Most small teams close the books late, by hand, on a Friday nobody wanted to work. Viktor does it before you ask. He is an AI employee that lives in Slack and Microsoft Teams and runs your month-end close on a schedule. He reconciles Stripe, QuickBooks, and your bank, catches duplicate charges and anomalies, disputes them, updates the runway model, and rebuilds the board deck. Sunday night it is done and posted to #finance. Monday morning you read it with coffee, already finished. Viktor works across 3,200+ integrations, keeps your data private (it never trains models on it), and is SOC 2 certified. Start free with $100 in credits, no credit card.
The AI Field: Anthropic released Claude Sonnet 5, its most capable mid-tier model yet, and made it the default for Free and Pro users. It's built for agents and gets near Opus-level performance for a fraction of the price.
Key details:
Performance is close to Opus 4.8 on reasoning, coding, and tool use, with a big jump in agentic tasks over Sonnet 4.6.
Intro pricing is $2 / $10 per million tokens through Aug 31, then $3 / $15.
It's the default model on Free and Pro, and available on Max, Team, and Enterprise.
Why This Matters: The race just shifted from "biggest model" to "cheapest capable agent." Putting near-frontier power in the free tier pressures everyone on price. Good enough, agentic, and cheap is the new sweet spot.
The AI Field: After nearly three weeks offline, Anthropic's most powerful model is returning. The US Commerce Department lifted the export controls, and Fable 5 (and Mythos 5) are being redeployed globally with new guardrails.
Key details:
Fable 5 is back today (July 1) on the Claude platform, Claude.ai, Claude Code, and Cowork, worldwide.
Anthropic trained a new safety classifier to block the specific cyber-jailbreak that triggered the ban.
Temporary cap: Pro, Max, Team, and Enterprise get Fable 5 for up to 50% of weekly limits through July 7, then via usage credits.
Why This Matters: The first government ban of an AI model is over in under a month, but it set a precedent: Washington can switch a frontier model off, and negotiate the terms of its return. The models are back; the leverage is new.
The AI Field: Meta unveiled Brain2Qwerty v2, an AI that reads your brain activity and turns it into typed text, no surgery required. It's a research milestone for people who can't communicate, not a gadget you'll buy.
Key details:
You wear a MEG scanner that reads brain signals as you try to type; the AI predicts the words, hitting 61% word accuracy (78% for the best participant).
It was trained on ~22,000 sentences from 9 volunteers, using large language models to fill in likely words from context.
The catch: MEG machines are massive, expensive lab equipment, so this stays in research for now.
Why This Matters: Non-invasive "mind typing" was science fiction a year ago. Even at 61%, crossing into usable accuracy without implants is a real breakthrough for people with paralysis. The link between brains and AI just got a lot closer.
🔥 THE AI FIELD'S TAKE
This week AI swung from scary back to useful and a little uncanny. For weeks the story was bans, cyberweapons, and national security. Now Anthropic drops a cheap, capable agent for everyone, quietly un-bans its most powerful model, and Meta shows a machine that reads your thoughts.
The through-line: AI is getting closer. Closer to free, closer to autonomous, and now, literally, closer to your brain.
The fear cycle cools. The "wait, this is actually wild" cycle is back.
OpenAI launched GPT-5.6 (Sol, Terra, Luna). A limited, government-aware preview to ~20 vetted orgs via API and Codex, with broad access "in the coming weeks."
Apple raised Mac and iPad prices 15-20%. It blamed an AI-driven memory-chip shortage (MacBook Air +$200); the stock fell about 6%.
Anthropic accused Alibaba of a massive distillation attack. It says 25,000 fake accounts generated 28.8M Claude interactions to train Alibaba's Qwen.
Google is bleeding AI talent. Nobel laureate John Jumper (AlphaFold) is joining Anthropic and Noam Shazeer is heading to OpenAI.
👀 WHAT TO WATCH
Fable 5 usage limits: the temporary 50% cap eases after July 7, then moves to usage credits. Watch how fast full access returns.
GPT-5.6 general availability: OpenAI says broad ChatGPT, Codex, and API access is "coming weeks."
Gemini 3.5 Pro: now expected in July after Google slipped its June deadline.
WRAP-UP
That's all for today's AI Brief.
💬 Claude Sonnet 5 is the new free default, near-Opus power for less.
🔓 Fable 5 is back online as the US lifts its export ban.
🧠 Meta's Brain2Qwerty types words straight from brain signals.
What did you think of today’s edition?
Until next time!
Olle | Founder of The AI Field