☕️ Anthropic leapfrogs OpenAI as the most valuable AI startup

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

💰 Anthropic leapfrogs OpenAI as the most valuable AI startup

🤖 Anthropic launches Claude Opus 4.8

💥 Blue Origin rocket explodes during ground test

🚐 Waymo launches Chinese-built Ojai robotaxi

🛑 Amazon shuts down internal AI leaderboard

Plus: 🎁 11 other news you might like, 🧰 6 tools, and 📚 4 papers.

💰 Anthropic leapfrogs OpenAI as the most valuable AI startup LINK

Anthropic has overtaken OpenAI as Silicon Valley's most valuable AI company after closing a $65 billion Series H round at a $965 billion valuation, edging past OpenAI's $852 billion mark from late March.

The round was led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, nearly tripling the $380 billion valuation from February and folding in $15 billion of prior commitments, including $5 billion from Amazon.

Anthropic's revenue run rate has climbed to $47 billion, up from $30 billion earlier this year, driven by Claude Code, and the company is quietly preparing its own IPO while OpenAI plans to file in the coming weeks.

🤖 Anthropic launches Claude Opus 4.8 LINK

Anthropic has released Claude Opus 4.8, a direct upgrade to Opus 4.7 that the company says improves coding, agentic tasks, reasoning, and knowledge work, available today across Claude endpoints and the API as claude-opus-4-8.

Regular pricing stays at $5 per million input tokens and $25 per million output tokens, while fast mode runs up to 2.5 times faster at $10 input and $50 output, three times cheaper than before.

Anthropic says Opus 4.8 is about four times less likely than Opus 4.7 to let flaws in its own generated code slip through without comment, and shows lower rates of misaligned behavior during agentic work.

💥 Blue Origin rocket explodes during ground test LINK

Blue Origin's New Glenn rocket blew up during a static fire test at Cape Canaveral, Florida on Thursday evening, as the company prepared for a fourth launch meant to carry Amazon Leo internet satellites to space.

Blue Origin confirmed an "anomaly" took place and said all personnel were accounted for, but didn't explain what went wrong, and the FAA, NASA, and Space Force have yet to respond to requests for comment.

The explosion will likely pause a program Jeff Bezos' company spent a decade building to compete with SpaceX, derailing plans for up to 12 New Glenn launches this year and its expected role in NASA's Artemis moon missions.

🚐 Waymo launches Chinese-built Ojai robotaxi LINK

Waymo has rolled out the Ojai, a Chinese-built electric minivan robotaxi made by Geely's Zeekr brand, offering free rides to select customers in Los Angeles, Phoenix, and San Francisco before a wider expansion.

The Ojai runs Waymo's sixth-generation system with 13 cameras, four lidar sensors, six radar units, and external audio receivers, and the modular setup also fits the Hyundai Ioniq 5 announced for the fleet.

Stripped-down Zeekr vehicles are shipped to Waymo's Arizona factory for outfitting, and the company is scaling toward tens of thousands of units a year to grow beyond its current fleet of 3,700 Jaguar I-Paces.

🛑 Amazon shuts down internal AI leaderboard LINK

Amazon has scrapped its internal "Kirorank" dashboard after staff gamed the AI leaderboard by pointing agents at pointless tasks to boost their scores, which ran up the company's cloud bills along the way.

The dashboard ranked employees based on activity on Amazon's Kiro developer platform, and Senior Vice President Dave Treadwell told staff, "Please don't use AI just for the sake of using AI," admitting it created extra costs.

Amazon will now track "normalized deployments", meaning AI-generated code that is actually useful, as the company pushes for more than 80 percent of developers to use AI weekly and plans roughly $200 billion in 2026 spending.

Other news you might like

• Mystery company accidentally blew $500 million on Claude in a single month — failed to put usage limit on licenses for employeesLINK

• Apple working to cram massive Gemini model into iPhone to power new SiriLINK

• Microsoft 365 Copilot gets a major redesign and performance boostLINK

• How long is Anthropic’s lease with SpaceX? Opinions vary.LINK

• Asana acquires no-code agent-builder Stack AILINK

• Euro-Office, Europe's open-source alternative to Microsoft Office and Google Docs, launches June 9LINK

• Fed up with vibe coders, dev sneaks data-nuking prompt injection into their codeLINK

• This company wants to clean your house for free, to train AI and robotsLINK

🧰 Trending tools

Ava 2.0: autonomous AI sales agent that sources leads from 250M+ contacts, runs multi-channel outreach, and books meetings without human oversight.LINK

/monitor by Firecrawl: instantly notifies your AI agent when specific web pages or sites change, keeping agents synced with live data without manual polling.LINK

Agent A by Ahrefs: an SEO toolset that simplifies keyword research, backlink analysis, and traffic growth for users at any experience level.LINK

MCP Bridge by Appfactor: automatically converts REST, GraphQL, SOAP, and gRPC APIs into typed MCP tool definitions, letting LLM agents access enterprise APIs through one standard interface.LINK

Ava Studio: automatically researches your product, generates 50+ short-form ad variants with hooks and creative angles, ready for TikTok, Reels, and Meta.LINK

Linear Diffs: a purpose-built product management tool designed for teams that need structured issue tracking, agent workflows, and streamlined planning in one place.LINK

📚 Trending papers & reports

AI propaganda factories show that persona design matters more than which AI model is used, and that pushing back on arguments actually makes the messaging more extreme and ideologically rigid.LINK

Cassandra, a new AI inference framework, speeds up reasoning AI models on consumer devices by up to 2.41x over the standard baseline, using smart pruning techniques that require no extra training.LINK

AutoScientists, a decentralized AI research system, outperforms existing AI agents across scientific tasks, achieving a mean leaderboard percentile of 74.4% across 24 biomedical tasks and finding 7 improvements where rival systems find none.LINK

Negation Neglect, a newly discovered flaw where fine-tuning AI models on documents warning a claim is false backfires, causing models to believe the claim is true, with belief rates jumping from 2.5% to 88.6%.LINK

Want to get the latest news differently? Find us on:

See you tomorrow for a new dose of ☕️ Techpresso!