In partnership with
Hello from The AI Night,
Today in AI:
Anthropic Announces SpaceX Deal, Doubles Claude Limits
OpenAI GPT-5.5 Instant Cuts Hallucinations 52%
Perplexity Launches Computer for Professional Finance Teams
Here's the deal: Anthropic signed an agreement with SpaceX to use all compute capacity at the Colossus 1 data center, adding over 300 megawatts and 220,000 NVIDIA GPUs within the month. The company also raised usage limits across Claude Code and its API.
The Breakdown:
Claude Code's five-hour rate limits are doubled for Pro, Max, Team, and seat-based Enterprise plans.
Peak hours limit reductions are removed for Pro and Max Claude Code users.
API rate limits for Claude Opus models increase considerably.
The deal joins prior agreements: 5 GW with Amazon, 5 GW with Google and Broadcom, $30 billion in Azure capacity, and $50 billion with Fluidstack.
Anthropic expressed interest in developing a multi-gigawatt orbital AI compute with SpaceX.
The bigger picture: Every other AI lab announces compute deals months before delivery. Anthropic signed with SpaceX and doubled user limits in the same breath. Two hundred twenty thousand GPUs arriving within weeks, not quarters. The orbital compute mention sounds like science fiction today. But so did every number in this paragraph a year ago.
Here's the deal: OpenAI released GPT-5.5 Instant as the new default ChatGPT model for all users, replacing GPT-5.3 Instant. The update targets accuracy, conciseness, and personalization across everyday tasks.
The Breakdown:
Hallucinated claims dropped 52.5% on high-stakes prompts in medicine, law, and finance. Inaccurate claims fell 37.3% on user-flagged conversations.
Benchmark scores jumped across GPQA (78.5% to 85.6%), AIME 2025 (65.4% to 81.2%), and MMMU-Pro (69.2% to 76.0%).
Responses use 30.2% fewer words with reduced overformatting and fewer unnecessary follow-up questions.
Personalization now pulls context from past chats, files, and connected Gmail for Plus and Pro users.
New memory sources feature shows what context shaped each response, with controls to delete or correct stored data.
GPT-5.3 Instant remains available for paid users for three months before retirement.
The bigger picture: OpenAI's most important model is not its smartest. It is the one 900 million people use weekly. Cutting hallucinations in half on medical and legal prompts is not a benchmark win. It is a liability fix enterprise buyers care about. OpenAI upgraded the free tier before the premium one. That tells you where the real battle is.
Here's the deal: Perplexity launched Computer for Professional Finance, bringing licensed data from Morningstar, PitchBook, Daloopa, and Carbon Arc into its AI interface. The product ships with 35 pre-built workflows for recurring analyst tasks.
The Breakdown:
Company tearsheets generate business summaries, key metrics, and recent news from a single company input.
Annotated stock charts mark tickers with earnings, product launches, and market catalysts.
Sourcing screens return filtered company lists with match reasoning and signal breakdowns.
Every output links citations to SEC filings, earnings transcripts, and licensed data sources.
The 35 workflows target tasks analysts repeat weekly, reducing manual data assembly.
The bigger picture: Finance analysts pay $25,000 a year for Bloomberg terminals and still copy numbers into spreadsheets by hand. Perplexity just built a tool that pulls licensed data and cites every figure back to its original SEC filing. Bloomberg's whole pitch has always been trust. Now someone else is selling it cheaper.
Write docs 4x faster. Without hating every second.
Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.
Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.
Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.
What else you need to know:
Vercel open sourced deepsec, a security scanning harness that uses Claude Opus 4.7 and GPT 5.5 coding agents to investigate codebases for vulnerabilities, running locally or across 1,000+ concurrent sandboxes.
OpenAI open-sourced MRC, a new GPU networking protocol co-developed with AMD, Broadcom, Intel, Microsoft, and NVIDIA, that sprays packets across hundreds of paths to eliminate congestion and route around failures in microseconds across 100,000+ GPU clusters.
Google's Gemini API now supports event-driven webhooks, replacing inefficient polling with real-time HTTP POST notifications when long-running tasks like batch processing or Deep Research complete.
EVE Online's parent company split from Pearl Abyss, rebranded as Fenris Creations, and partnered with Google DeepMind to research general-purpose AI in EVE's player-driven virtual environment.
xAI launched Quality Mode for its Grok Imagine API, adding improved realism, multilingual text rendering, and tighter prompt following, now ranking third on LMArena's text-to-image leaderboard.
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.
2 ways 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