# OpenAI Lands Noam Shazeer as Life-Science AI Advances and Frontier Costs Rise

*By AI High Signal Digest • June 18, 2026*

OpenAI made the day’s biggest talent move by hiring Noam Shazeer while also pushing deeper into life-science AI with a new benchmark and a lab-validated chemistry result. Meanwhile, Claude Fable 5 reset the price curve for frontier models, and G7 policy discussions moved toward tighter control over model access and hardware.

## Top Stories

*Why it matters: leadership changes, scientific validation, and frontier-model economics all shifted in ways that will shape the next phase of AI competition.*

- **OpenAI hired Noam Shazeer from Google DeepMind.** Shazeer is leaving his VP Engineering / Gemini co-lead role at Google DeepMind to join OpenAI, where the company said he will serve as **lead for architecture research** [^1][^2]. He said the move was a difficult decision after work he was proud of at Google [^3].
- **OpenAI pushed deeper into life-science AI.** It introduced **LifeSciBench**, built with **173 scientists** and **750 expert-authored tasks** across seven biological research workflows [^4]. OpenAI said **GPT-Rosalind** scored above GPT-5.5 across all seven workflows [^5]. Separately, OpenAI said **GPT-5.4** helped drive a medicinal chemistry project to a validated experimental result, with improved yields for **88%** of boronic acids and **83%** of sulfonamides tested [^6][^7].
- **Claude Fable 5 raised the cost of the frontier.** Artificial Analysis reported **Fable 5** at **60** on its Intelligence Index, ahead of Claude Opus 4.8 at 56 and GPT-5.5 at 55 [^8]. It also reported list pricing of **$10/$50** per 1M input/output tokens and about **$6.2K** to run the benchmark suite, its highest recorded benchmark cost [^9][^8].

## Research & Innovation

*Why it matters: today’s most useful technical work focused on making agents faster, measuring them in more realistic settings, and extending AI into longitudinal healthcare.*

- **PreAct** turns a computer-use agent’s first successful run into a replayable state-machine program, then reuses it without per-step model calls for **8.5x to 13x** faster execution; if the screen no longer matches, control returns to the agent [^10].
- **iOSWorld** released a benchmark for personally intelligent phone agents across **26 custom iOS apps** and **133 tasks**. Even with privileged vision+XML access, the strongest frontier model reached only **52%** success [^11].
- **Google’s AMIE** moved from one-off diagnosis toward longitudinal disease management. In a multi-visit study with patient actors, Google said it reached physician-level performance and scored higher on plan preciseness and guideline alignment [^12][^13].

## Products & Launches

*Why it matters: the developer stack is getting more agent-native, with tighter orchestration, routing, and design-to-code loops.*

- **GitHub Copilot app** is now generally available as a home base to pick up tasks, direct agents in parallel, and land PRs [^14]. GitHub also said Copilot’s **Auto mode** now uses a routing model that weighs reasoning depth, code complexity, debugging difficulty, and tool orchestration needs [^15].
- **Cursor** added cloud subagents. Users can launch a subagent in its own cloud VM with `/in-cloud`, keep environments as reusable snapshots, and continue prompting from a phone while agents work in parallel [^16][^17][^18].
- **Claude Design** now syncs both ways with **Claude Code**. Anthropic said `/design-sync` can pull a design system into a repo or push builds back to the canvas for further editing [^19][^20].

## Industry Moves

*Why it matters: companies are moving from pilot demos to capital allocation, internal deployment, and ecosystem bets.*

- **Block** said its internal **Builderbot** now handles **200,000 operations per day**, merges **1,500 pull requests per week**, and is responsible for **15%** of production code changes; it said work that used to take months now takes days [^21].
- **XDOF** announced a **$70M** raise to build infrastructure for robot foundation models and said it is open-sourcing **ABC-130K**, described as the largest open-source teleoperation dataset [^22].
- **OpenAI** committed **$600,000** to the Rust Foundation and said it is continuing to bet on Rust as the future of systems programming [^23][^24].

## Policy & Regulation

*Why it matters: AI governance is shifting from abstract debate to concrete controls over access, hardware, and model release conditions.*

- At the **G7**, Dario Amodei and Demis Hassabis called for a **U.S.-led coalition** to set AI standards and rules; reporting said the proposal included structuring access to frontier models and hardware in a way that excludes China [^25][^26].
- U.S. officials told **WIRED** that Anthropic would need to ensure **Fable 5** guardrails cannot be circumvented before any rerelease; the same report said security experts do not think that can be done [^27].

## Quick Takes

*Why it matters: these smaller updates still show where new capability and infrastructure are appearing next.*

- Midjourney announced a new division called **Midjourney Medical** and shared a technical dive into its **Midjourney Scanner** [^28][^29].
- **Grok Imagine Video 1.5** launched with sharper realism, better physics, and faster generations [^30].
- Google Cloud introduced the **Open Knowledge Format**, a markdown-and-YAML spec for AI context, and said Knowledge Catalog can ingest it natively [^31].
- **GLM-5.2** is now available on Together AI for long-context, tool-heavy agent workloads [^32][^33].

---

### Sources

[^1]: [𝕏 post by @scaling01](https://x.com/scaling01/status/2067407207956705517)
[^2]: [𝕏 post by @scaling01](https://x.com/scaling01/status/2067410385448120488)
[^3]: [𝕏 post by @NoamShazeer](https://x.com/NoamShazeer/status/2067400851438932297)
[^4]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2067346916929937827)
[^5]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2067346918746030496)
[^6]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2067293745075442171)
[^7]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2067293749412339980)
[^8]: [𝕏 post by @ArtificialAnlys](https://x.com/ArtificialAnlys/status/2067384319942029379)
[^9]: [𝕏 post by @ArtificialAnlys](https://x.com/ArtificialAnlys/status/2067384327424585812)
[^10]: [𝕏 post by @dair_ai](https://x.com/dair_ai/status/2067386816815387019)
[^11]: [𝕏 post by @rsalakhu](https://x.com/rsalakhu/status/2064402156740907444)
[^12]: [𝕏 post by @GoogleForHealth](https://x.com/GoogleForHealth/status/2067290472599462187)
[^13]: [𝕏 post by @GoogleResearch](https://x.com/GoogleResearch/status/2067389215239999488)
[^14]: [𝕏 post by @github](https://x.com/github/status/2067281833855709530)
[^15]: [𝕏 post by @pierceboggan](https://x.com/pierceboggan/status/2067449111809597803)
[^16]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2067378592938418422)
[^17]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2067378591101235666)
[^18]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2067366343817805899)
[^19]: [𝕏 post by @claudeai](https://x.com/claudeai/status/2067325887909884315)
[^20]: [𝕏 post by @ClaudeDevs](https://x.com/ClaudeDevs/status/2067391951725629941)
[^21]: [𝕏 post by @blocks](https://x.com/blocks/status/2067284573482815979)
[^22]: [𝕏 post by @philippswu](https://x.com/philippswu/status/2067277090010136949)
[^23]: [𝕏 post by @charliermarsh](https://x.com/charliermarsh/status/2067280638994968657)
[^24]: [𝕏 post by @OpenAIDevs](https://x.com/OpenAIDevs/status/2067280779689042235)
[^25]: [𝕏 post by @AndrewCurran_](https://x.com/AndrewCurran_/status/2067305163556364497)
[^26]: [𝕏 post by @MTSlive](https://x.com/MTSlive/status/2067306564747395346)
[^27]: [𝕏 post by @WIRED](https://x.com/WIRED/status/2067292536687804692)
[^28]: [𝕏 post by @midjourney](https://x.com/midjourney/status/2067421950314688759)
[^29]: [𝕏 post by @midjourney](https://x.com/midjourney/status/2067422898407837797)
[^30]: [𝕏 post by @xai](https://x.com/xai/status/2067092897951109427)
[^31]: [𝕏 post by @GoogleCloudTech](https://x.com/GoogleCloudTech/status/2067012903337664886)
[^32]: [𝕏 post by @togethercompute](https://x.com/togethercompute/status/2067435704313930234)
[^33]: [𝕏 post by @togethercompute](https://x.com/togethercompute/status/2067435708256571497)