AI Daily Dev — June 30, 2026

Meta's Brain2Qwerty v2 Decodes Typed Sentences From MEG With 78% Best-Subject Accuracy

Non-invasive brain-to-text pipeline reaches 61% mean word accuracy across 9 volunteers — versus the prior 8% ceiling for non-implant methods.
Best participant hits 78% word accuracy with more than half of sentences decoded at one word error or less.
Trained on ~22,000 sentences, 10 hours per participant in a magnetoencephalography (MEG) scanner.
Meta released full v1 and v2 training code; the Basque Center on Cognition, Brain, and Language released the v1 dataset.

research ai.meta.com

Anthropic MD Chris Ciauri: 'We are very confident the models will become available again in the coming days.'
Polymarket 'restored by June 30' contract sits at 41.5% (was 9.5% on June 26) as Mythos cleared earlier in the week.
Tom Brown replaced Dario Amodei in daily Commerce talks; Trump said on June 19 that Anthropic is 'no longer a national-security threat'.
Leaked Anthropic code hints at gated US-first restoration with weekly usage caps and government-ID verification (live July 8).

industry axios.com

Day 2 of the 6,000-engineer Moscone event opens with the Coding Agents keynote across 10 tracks.
Anthropic unveiled the official MCP Registry API — the central index now holds ~2,000 server entries, up 407% from the September 2025 preview.
Mahesh Murag (Anthropic) presenting Claude skills; Tom Dohmke speaking at the GitHub HQ side event Tuesday evening.
Agenda dominated by MCP, verifier-driven loops, and 'agent reliability' — the topics that have crowded out vibe-coding on dev Twitter.

community ai.engineer

Ramp's May AI Index: 34.4% of US firms now pay for Anthropic tools vs 32.3% for OpenAI — first time Anthropic leads.
Anthropic credits Claude Code, then says 65% of its own product team's commits are now authored by Claude Tag (their internal Slack agent).
Claude Tag (launched June 23) replaces the old Claude-in-Slack app and turns Claude into a shared, ambient teammate inside a channel.
Anthropic May run-rate revenue crossed $47B — context for the $965B Series H two months ago.

industry venturebeat.com

Reddit/X threads show monthly Copilot bills jumping from $29 to ~$750, and from $50 to $3,000 on heavy agentic workflows.
Switch from flat-rate to token-based 'AI Credits' took effect June 1 — 30 days in, dev forums are still dominated by migration threads.
Cursor's Standard/Premium seat split and Devin Desktop's $20-$200 tiering are the most-cited alternatives; Claude Code usage spikes.
Microsoft's pitch — 'use Copilot for everything' — built the workflows that metered pricing now charges 10–50× more for.

tools theregister.com

V4.1 ships native MCP protocol support (no adapter layer), image + audio multimodal input, and enterprise tool-chain integration.
Two months after V4 (April 24); Hugging Face has it as the top trending model within a week of release.
Sixth open-weight Chinese frontier release of the month — joins Qwen 3.6, Hunyuan Large 3, ERNIE 5.1, Doubao Pro, GLM 5.2.
Marks DeepSeek's pivot from 'no commercialization' to enterprise sales, funded by a recent 5B yuan round.

open-source deepseek.com