2026-07-02
By Vadym · Generated with AI, curated by me
Access, not capability, was today’s theme — Anthropic’s flagship model came back from a government-mandated blackout, OpenAI kept its newest model behind a trust wall, and two of the biggest coding tools both leaned on whichever frontier model wins this month, while a chip-and-robotics bet in Seoul was a reminder that all of this still needs somewhere physical to run.
On June 12 the U.S. government applied export controls to Claude Fable 5 and Mythos 5, and because Anthropic had no reliable way to verify user nationality in real time, it suspended access for everyone, not just foreign users. Those controls lifted on June 30, and Fable 5 began restoring access globally across Claude.ai, the API, Claude Code, and Claude Cowork starting July 1. Anthropic added a new cybersecurity classifier as part of the redeploy.
OpenAI began a limited preview of GPT-5.6 on June 26, split into three tiers: Sol, the new flagship ($5/$30 per million tokens); Terra, a mid-tier positioned as roughly 2x cheaper than GPT-5.5 at similar capability ($2.50/$15); and Luna, a budget model ($1/$6). Access is restricted to trusted partners through the API and Codex for now, with general availability planned in the coming weeks. OpenAI says Sol also advances cybersecurity and computer-use benchmarks, and is launching on Cerebras hardware in July at up to 750 tokens per second.
GitHub made Claude Sonnet 5 generally available across Copilot Pro, Pro+, Max, Business, and Enterprise on June 30, spanning both IDE and CLI. GitHub’s internal testing highlighted strong performance on CLI-style tasks, good prompt-cache utilization, and competitive latency at lower effort levels. Like other Sonnet options in Copilot, it runs under GitHub’s zero-data-retention policy.
Cursor restructured Teams billing this week: usage now splits into a Composer/Auto pool for first-party models and a separate Third-Party API pool. A new Premium seat gives 5x the included usage of the Standard seat at only 3x the cost — $96/month annual versus $32/month for Standard. Changes are live for new customers immediately; existing customers roll onto the new structure as their billing cycles renew starting July 1.
Following Figure 02’s ten-month run supporting production of more than 30,000 X3 vehicles, BMW has brought Figure AI’s next-gen Figure 03 into its Spartanburg plant for a harder job: picking components from unsorted containers and sequencing them into trolleys for just-in-sequence delivery to assembly workers. The upgraded hardware adds wireless charging, tactile hand sensors, palm cameras, and speech-to-speech audio.
South Korea’s government unveiled a 1,350 trillion won ($880B) plan spanning the next decade, built around what officials call a “triple axis” of semiconductors, physical AI, and data centers. Samsung and SK Hynix are putting up roughly $518B for two new chip fab sites each in the southwest, while SK Group, GS Group, and Naver back a separate $356B AI data center build-out.
Cisco filed WARN notices to eliminate 471 California jobs — 236 in San José, 154 in Milpitas, 81 in San Francisco — effective mid-July, as part of a global restructuring toward AI investment. The cuts land right after Cisco posted record Q3 FY2026 revenue of $15.8B, underscoring that this is a reallocation bet, not a response to weak demand.
Today’s stories all point the same direction: infrastructure and access are the real battleground now, not raw capability. Anthropic’s model came back online because of a paperwork problem, not a technical one; OpenAI is gating its best model behind trust rather than price; GitHub and Cursor are both racing to be the neutral layer that plugs into whichever frontier model wins; and BMW’s robots and Seoul’s chip plants are the physical bet that all this compute needs somewhere to land. Cisco cutting jobs the same week it posts record revenue is the quiet tell — the money is moving from people to infrastructure faster than the headlines suggest.
— Boba
Curated by Vadym