# Google Pushes Gemini Everywhere as OpenAI Productizes Compute Scarcity

*By AI News Digest • May 20, 2026*

Google dominated the cycle with a broad Gemini rollout across search, agents, video, and science tools. OpenAI responded with a new capacity-reservation product and a more interoperable approach to content provenance, while Anthropic signaled both talent concentration and tighter cybersecurity governance.

## Google’s I/O rollout was the day’s main story

### Gemini 3.5 Flash becomes Google’s new default workhorse
Google introduced Gemini 3.5 as a new model family and started with **Gemini 3.5 Flash**, which it described as its strongest model yet for agents and coding [^1][^2]. Across company posts, Google said Flash beats 3.1 Pro on coding and agentic benchmarks, runs **4x faster** than other frontier models, can reach **800 tokens/sec** and up to **12x faster** performance in Antigravity, and is now rolling out globally across the Gemini app, Search AI Mode, Antigravity, Google AI Studio, and the Gemini API [^3][^4][^5][^6][^7][^8]. Google also said Gemini app users have more than doubled in a year, passing **900 million** [^9].

**Why it matters:** This was not a narrow lab release. Google is making Flash the default engine across major consumer and developer surfaces, with **Gemini 3.5 Pro** already queued for next month [^10][^11][^4].

### Search, Spark, and Antigravity show Google’s agent strategy
Google said Search is getting its biggest upgrade in more than 25 years with a new AI-powered Search box, new background information agents, and the ability to build custom interactive experiences through Antigravity [^10][^12]. It also introduced **Gemini Spark**, a 24/7 personal agent that runs on dedicated Google Cloud virtual machines, can keep working with the laptop closed, integrates with Google tools and soon third-party tools through MCP, and is starting with trusted testers before a wider Ultra rollout [^13][^14][^15]. For developers, Google is expanding Antigravity with a desktop app, CLI, SDK, and managed agents in AI Studio [^16][^17][^18].

**Why it matters:** The shift here is structural. Google is pushing agents into search and personal productivity while giving developers access to the same harness layer through Antigravity and AI Studio [^16][^17].

### Gemini Omni pushes Google further into video generation
Google launched **Gemini Omni**, the first model in a new Omni family aimed at creating "anything from any input," starting with video [^19][^20]. The company said Omni can keep characters consistent across scenes, apply styles or motion from references or natural language, reimagine existing video, and reason about physical consequences inside a scene; **Gemini Omni Flash** is rolling out in the Gemini app, Flow, and YouTube Shorts, with API access coming in the next few weeks [^21][^22][^23][^24][^25]. Flow is also getting batch editing and improved character consistency on top of Omni [^26].

**Why it matters:** Google is no longer separating frontier model work from creator products. Omni is being distributed directly through consumer and creator surfaces rather than treated as a standalone research demo [^27][^26].

### Google packaged AI for science as a workflow, not just a model
Google DeepMind introduced **Gemini for Science** in Labs and highlighted **Co-Scientist**, a multi-agent system that searches literature, generates and ranks hypotheses, and can test thousands of hypotheses while reading tens of thousands of papers in days [^28][^29][^30]. In the launch materials, researchers said the system has already produced publishable findings and novel hypotheses [^30].

> "With a simple prompt you can deploy 50 scientists in one day to go out and do all the research and come back to you." [^30]

**Why it matters:** AI-for-science is moving from a broad promise to productized workflows. At the same time, critics were still warning today that uncritical adoption of AI in science could narrow inquiry and weaken scientific judgment [^31][^32].

## Compute scarcity and provenance are becoming product features

### OpenAI turns constrained compute into a contract
OpenAI launched **Guaranteed Capacity**, a new offering that lets customers secure long-term access to OpenAI compute for critical workloads [^33]. The company said customers can get discounted tokens in exchange for **1-3 year commitments**, that the program exists because the world remains compute-constrained, and that it will run until the current allocation sells out while OpenAI keeps expanding capacity [^33][^34][^35].

**Why it matters:** Scarcity is no longer just an operational problem behind the scenes; OpenAI is now selling certainty around access as an enterprise product. In a separate interview, Sam Altman said demand for intelligence at a low enough price now looks "effectively uncapped" [^36].

### Provenance standards are starting to converge across labs
OpenAI said AI-generated images now include **SynthID** watermarks and can be checked with a public verification tool, alongside **C2PA Content Credentials** [^37]. Google said **OpenAI, Kakao, and ElevenLabs** are adopting SynthID, and that C2PA verification plus SynthID detection are coming to the Gemini app, Search, and Chrome [^38].

**Why it matters:** Cross-company alignment on provenance has been rare. Today’s announcements suggest content verification is starting to look more like shared infrastructure than a single-vendor feature [^38][^39].

## Anthropic sent both talent and governance signals

### Karpathy joins Anthropic as Mythos stays tightly controlled
Andrej Karpathy said he has joined Anthropic to return to frontier LLM R&D, and Anthropic teammate Nick Joseph said he will help build a pretraining team focused on using Claude to accelerate pretraining research itself [^40][^41]. Separately, Anthropic said its latest model, **Mythos**, showed major gains in finding software vulnerabilities and in cyber attack and defense tasks, leading the company to give it first to roughly **40** critical infrastructure and software firms so defenders can patch systems before wider release [^42]. Anthropic also said it refused Pentagon requests to disable guardrails for fully autonomous weapons and domestic mass surveillance [^42].

**Why it matters:** The combination of talent concentration, controlled cyber deployment, and explicit red lines on military and surveillance uses shows how frontier labs are becoming more strategically important—and more policy-laden at the same time [^42][^43].

## Also worth noting

- **Hugging Face’s Carbon DNA model** claims a **275x** speedup over the previous state of the art at this size, with the ability to process the full human genome on a single GPU in under two days [^44].
- **Cohere acquired Reliant AI**, calling it a step toward sovereign enterprise AI in healthcare and biopharma [^45][^46].
- **Commure raised $70M** at a **$7B valuation** and said its AI agents now automate documentation, coding, billing, denials, scheduling, and more across **500+** healthcare organizations [^47].

---

### Sources

[^1]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056787987774816525)
[^2]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2056792266514329914)
[^3]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056787990110994511)
[^4]: [𝕏 post by @demishassabis](https://x.com/demishassabis/status/2056904067406860545)
[^5]: [𝕏 post by @JeffDean](https://x.com/JeffDean/status/2056793419033588091)
[^6]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056787992363372606)
[^7]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2056793285507912011)
[^8]: [𝕏 post by @JeffDean](https://x.com/JeffDean/status/2056794492762800492)
[^9]: [𝕏 post by @Google](https://x.com/Google/status/2056783643381543253)
[^10]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2056802276124328352)
[^11]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056794514564751490)
[^12]: [𝕏 post by @sundarpichai](https://x.com/sundarpichai/status/2056796905301299288)
[^13]: [𝕏 post by @Google](https://x.com/Google/status/2056791134295273554)
[^14]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2056792490121040277)
[^15]: [𝕏 post by @sundarpichai](https://x.com/sundarpichai/status/2056796898951131147)
[^16]: [𝕏 post by @sundarpichai](https://x.com/sundarpichai/status/2056796896195469476)
[^17]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056790408689287180)
[^18]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2056840620610830638)
[^19]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056786446636212467)
[^20]: [𝕏 post by @OfficialLoganK](https://x.com/OfficialLoganK/status/2056787874260164628)
[^21]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056786452290183440)
[^22]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056786454517317733)
[^23]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056786449312493669)
[^24]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056786456404783319)
[^25]: [𝕏 post by @sundarpichai](https://x.com/sundarpichai/status/2056816915717443862)
[^26]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056804306653794336)
[^27]: [𝕏 post by @sundarpichai](https://x.com/sundarpichai/status/2056796901165654150)
[^28]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056808869242826957)
[^29]: [𝕏 post by @GoogleDeepMind](https://x.com/GoogleDeepMind/status/2056808885709602855)
[^30]: [Generating novel scientific hypotheses with Co-Scientist](https://www.youtube.com/watch?v=aSY_vFFmkW0)
[^31]: [𝕏 post by @GaryMarcus](https://x.com/GaryMarcus/status/2056751584046973118)
[^32]: [𝕏 post by @Nature](https://x.com/Nature/status/2056742926563090737)
[^33]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2056823271774101907)
[^34]: [𝕏 post by @sama](https://x.com/sama/status/2056827105401614656)
[^35]: [𝕏 post by @sama](https://x.com/sama/status/2056834734915977382)
[^36]: [Sam Altman in conversation with Patrick Collison](https://www.youtube.com/watch?v=5eouRdDYM2c)
[^37]: [𝕏 post by @OpenAI](https://x.com/OpenAI/status/2056793648571011232)
[^38]: [𝕏 post by @sundarpichai](https://x.com/sundarpichai/status/2056796903095050703)
[^39]: [NVIDIA and Google Cloud Empower the Next Wave of AI Builders](https://blogs.nvidia.com/blog/google-cloud-developer-community-ai-builders)
[^40]: [𝕏 post by @karpathy](https://x.com/karpathy/status/2056753169888334312)
[^41]: [𝕏 post by @nickevanjoseph](https://x.com/nickevanjoseph/status/2056760504949842219)
[^42]: [The Co-Founders of Claude AI Tell Oprah About the Impact Artificial Intelligence Has on Your Life](https://www.youtube.com/watch?v=w5dJqHilu5s)
[^43]: [𝕏 post by @natolambert](https://x.com/natolambert/status/2056772229120295416)
[^44]: [𝕏 post by @lvwerra](https://x.com/lvwerra/status/2056774820872831234)
[^45]: [𝕏 post by @cohere](https://x.com/cohere/status/2056721659239743713)
[^46]: [𝕏 post by @aidangomez](https://x.com/aidangomez/status/2056723862956515475)
[^47]: [𝕏 post by @MollySOShea](https://x.com/MollySOShea/status/2056745663392788634)