In partnership with
Hello from The AI Night,
Today in AI:
Anthropic Hits $965B Valuation in Series H
OpenAI Launches Rosalind Biodefense for Trusted Developers
Xiaomi Cuts MiMo API Pricing By 99%
Here's the deal: Anthropic closed a $65 billion Series H round at a $965 billion post-money valuation. The round was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital. Run-rate revenue crossed $47 billion.
The Breakdown:
Round led by Altimeter, Dragoneer, Greenoaks, and Sequoia, with Capital Group, Coatue, GIC, and ICONIQ co-leading.
Includes $15 billion of previously committed hyperscaler investment, with $5 billion from Amazon.
Memory and chip suppliers Micron, Samsung, and SK hynix joined as strategic infrastructure partners.
Signed Amazon for up to five gigawatts of new capacity and Google and Broadcom for five gigawatts of TPU capacity.
SpaceX agreement provides GPU access in Colossus 1 and Colossus 2.
Claude is the first frontier model on AWS, Google Cloud, and Microsoft Azure, with AWS as primary training partner.
The bigger picture: Two years ago Anthropic was the safety-focused underdog. Now it approaches a trillion-dollar valuation with every major cloud fighting to host it. No other AI lab runs natively on AWS, Google Cloud, and Azure simultaneously. That makes Anthropic impossible to cut off and easy to buy from. When your infrastructure partners are also your competitors, being everywhere is the only safe position.
Here's the deal: OpenAI launched Rosalind Biodefense, a program giving vetted developers access to GPT-Rosalind, its frontier reasoning model built for life sciences research. The company also expanded trusted access for select U.S. government and allied partners working on public health and biodefense missions.
The Breakdown:
GPT-Rosalind is OpenAI's frontier reasoning model built for life sciences research.
OpenAI sponsors model access and provides launch support to trusted developers building biodefense applications.
Initial developer cohort includes Fourth Eon, SecureDNA, SecureBio Detection, and ProEquip.
Government partners include Lawrence Livermore and Johns Hopkins Applied Physics Laboratory, plus CEPI for vaccine work.
Use cases span epidemiological modeling, early detection, screening, and medical countermeasure development.
OpenAI first treated July 2025's ChatGPT agent as High Capability in biology under its Preparedness Framework.
The bigger picture: OpenAI flagged this model as high-risk under its own safety framework, then gave it to biodefense labs anyway. That is the new playbook for dual-use AI. Powerful enough to be dangerous, distributed only to defenders. Lawrence Livermore and Johns Hopkins are not startup partners. They are the institutions governments call during actual crises. If gated access works here, it becomes the template for every dangerous capability that follows.
Here's the deal: Xiaomi permanently reduced MiMo-V2.5 API pricing by up to 99% versus previous rates, effective May 26 at 6:00 PM PDT. The company also unified pricing across all context lengths and upgraded its Token Plans. It attributed the cuts to continued inference and serving efficiency gains across the MiMo stack.
The Breakdown:
API pricing cut up to 99%, with unified rates applied across all context lengths.
Token Plans now deliver 5 to 8 times more usable tokens at the same price.
Billing rules were simplified into more transparent usage terms.
All current Token Plan credits are being fully reset as a thank-you to existing users.
MiMo-V2.5-TTS remains free for a limited time.
The 100T Token Grant for Builders concluded early, with all 100 trillion tokens claimed since its April 28 launch.
The Apache Software Foundation committers benefit program continues long-term.
The bigger picture: Nobody cuts prices 99% because efficiency improved. They cut because nobody was buying. DeepSeek and Qwen ate the market on near-free inference while Xiaomi charged premium rates. Resetting every customer's credits is the tell. The old pricing was so wrong that wiping the slate cost less than honoring it. Xiaomi is not joining a price war. It is conceding one.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
What else you need to know:
xAI released grok-build-0.1 in public beta via its API, an agentic coding model priced at $1 per million input and $2 per million output tokens, also available through OpenRouter, Cursor, and OpenCode.
Google introduced Flow Agent, a Gemini-powered creative assistant that helps plan and reason through video projects, suggests dialogue and plot ideas, generates multiple scene variations at once, and organizes final assets into collections.
OpenAI added Windows support to Codex, enabling computer use that takes action on Windows machines, plus task control through the ChatGPT mobile app so users can start, review, and steer work remotely.
NVIDIA's Vera CPU beat AMD's EPYC 9575F by 10% in Phoronix benchmarks, delivering 88 Olympus cores, 1.2TB/s memory bandwidth, and a 1.6x geometric mean gain over Grace.
ElevenLabs launched Music v2, powering ElevenMusic, ElevenCreative, and the upcoming ElevenAPI, with mid-track genre transitions, improved inpainting, licensed training data, and self-serve pricing cuts up to 50%.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
2ways to support us:
Forward this to your AI-curious friend → https://www.theainight.com
Sponsor The AI Night and reach 500+ AI builders daily → passionfroot.me/theainight