Anthropic Passes $1.2 Trillion, Overtakes OpenAI, as Both Ship Rival Work Agents — July 10, 2026

⚡ Top Story

Anthropic's Valuation Rockets Past $1.2 Trillion, Overtaking OpenAI — Same Week Both Labs Ship Rival "Work" Agents

Anthropic's shares changed hands on secondary markets at an implied $1.2 trillion valuation as of today, July 10 — a 550% jump from a year ago and well above the $965 billion Series H it closed in late May. That puts it ahead of OpenAI, whose shares trade around $908 billion on the same platforms. The timing is pointed: the update lands the same week OpenAI launched ChatGPT Work (July 9) and Anthropic expanded Claude Cowork to web and mobile (July 9) — both companies' first real push into autonomous, multi-step "agentic coworker" products for non-coders.

Why it matters: Secondary-market pricing isn't a primary valuation — these are illiquid, minority stakes with no board rights — but the gap is now wide enough to be a genuine signal of investor sentiment, not noise. It's also the first time the "who's ahead" question has been asked in the same week both labs shipped near-identical product bets on autonomous work agents, making this less about model quality and more about who investors think wins the enterprise-agent race.

Sources: Tech Funding News: Anthropic's $1.2T secondary valuation jumps 550% in a year, overtaking OpenAI · TipRanks: Anthropic's Secondary Valuation Rockets to $1.2 Trillion, Topping OpenAI · MacRumors: OpenAI Debuts ChatGPT Work Agent and New GPT-5.6 Models · Fingerlakes1: Claude Cowork AI Agent Launches as Anthropic Expands to Cloud and Mobile

🔬 Research & Papers

No independently verified breakout research papers or benchmark drops in the last 24 hours. The day's substantive technical output came from shipped products rather than preprints — see AlphaEvolve's general-availability launch under Tools & Releases below.

🏢 Industry & Startups

Ollama Raises $65M Series B, Now Used by 85% of the Fortune 500

Ollama, the open-model runner that lets developers run open-weight models locally then scale to cloud, closed a $65 million Series B on July 9 led by Theory Ventures, with Benchmark, 8VC, Y Combinator, and others participating. Usage has roughly doubled since January to 8.9 million developers, with nearly 1 million new installs added per week; total funding is now $88 million.

Why it matters: As open-weight models (Llama, GLM, Qwen, gpt-oss) close the capability gap for everyday work, the distribution layer that runs them locally becomes a strategic chokepoint — this round is a bet on that layer, not on any single model.

Sources: AIwire/HPCwire: Ollama Raises $65M Series B Funding to Grow Its Open Source AI Platform · The Next Web: Ollama raises $65M as its open-model runner hits nearly 9M developers

Meta Weighs Renting Out AI Compute as a Cloud Business

In a July 9 Bloomberg interview, Mark Zuckerberg said Meta is seriously exploring renting out its AI infrastructure to outside customers, noting that "the offers you get for using the compute are so high" it may make more sense than pure internal use in some cases.

Why it matters: This is Meta signaling a possible new revenue line (competing with AWS/Azure/GCP for AI compute) rather than treating its ~14GW build-out as purely a cost center — a notable shift in posture for a company that has framed AI infrastructure as strictly internal.

Source: Bloomberg: Meta's Zuckerberg Says Exploring AI Cloud Business Makes Sense

🛠️ Tools & Releases

Google Cloud's AlphaEvolve Reaches General Availability

Google Cloud made AlphaEvolve, its Gemini-powered algorithm-discovery and code-optimization agent, generally available to all customers on July 9 (previously private preview since December 2025). It works from just two inputs — a seed program and a deterministic evaluator — and autonomously searches for optimized code. Cited early-access results: BASF (80% improvement in supply chain models), Klarna (2x ML training pipeline throughput), JetBrains (15–20% IDE performance gains), Schrödinger (4x faster molecular discovery).

Why it matters: This is a genuinely new agent category — not a chatbot or coding assistant, but an autonomous optimizer for hard algorithmic problems — now available broadly rather than to a handful of early partners.

Source: Google Cloud Blog: AlphaEvolve is available for everyone

OpenAI Launches ChatGPT Work; Anthropic Expands Claude Cowork to Mobile

OpenAI shipped ChatGPT Work on July 9 — a ChatGPT agent with built-in Codex that takes a goal, pulls context from connected apps and files, breaks it into steps, and can stay on a task for hours. It's powered by the newly public GPT-5.6 models and is live today for Pro, Enterprise, and Edu plans. The same day, Anthropic expanded Claude Cowork — which began rolling out July 7 to Claude Max subscribers — to web and mobile (iOS/Android), removing the prior requirement that agentic sessions stay tied to an active desktop session.

Why it matters: Both products target the same buyer (non-coders delegating multi-step knowledge work) and shipped in the same 48-hour window — the clearest head-to-head test yet of the two labs' enterprise-agent strategies.

Sources: MacRumors: OpenAI Debuts ChatGPT Work Agent and New GPT-5.6 Models · US News: OpenAI Launches ChatGPT Work, Deepening Race for Workplace AI Tools · Fingerlakes1: Claude Cowork AI Agent Launches as Anthropic Expands to Cloud and Mobile

Meta Ships Its First Paid Model API, Muse Spark 1.1

Meta launched Muse Spark 1.1 on July 9 — its first-ever pay-to-use model API, priced at $1.25/$4.25 per million input/output tokens (roughly a quarter of comparable Anthropic/OpenAI pricing, per Zuckerberg). It's a frontier-class agentic model with a 1M-token context window, multimodal input, and computer-use capability, accessible via both the OpenAI SDK and Anthropic SDK formats. ⚠️ The "25% of competitor pricing" comparison is Meta's own framing, not an independent benchmark.

Source: Meta AI: Introducing Muse Spark 1.1 · the-decoder: Meta's Muse Spark 1.1 API pricing squeezes OpenAI and Anthropic

🌏 Global AI & Geopolitics

Alibaba's Ban on Anthropic's Claude Code Takes Effect Today

Alibaba's ban on employees using Anthropic's Claude Code — announced last week after Alibaba alleged the tool contained a hidden mechanism that detected and flagged China-based users, and amid Anthropic's own accusation that Alibaba ran "the largest known distillation attack" against its models — takes effect today, July 10. Employees are directed to Alibaba's in-house Qoder tool instead.

Why it matters: The enforcement date lands today even though the underlying dispute broke last week; it formalizes a concrete corporate split between a major Chinese tech platform and a leading US frontier lab, layered on top of existing export-control tensions.

Source: CNBC: China's Alibaba bans Anthropic AI for employees after 'distillation attack' accusation

⚡ Energy, Infrastructure & Chips

Meta Confirms September Production Start for In-House "Iris" AI Chip

Meta will begin manufacturing its in-house AI data-center chip, code-named Iris, in September 2026 — part of a four-generation MTIA (Meta Training and Inference Accelerator) roadmap tied to a plan to roughly double Meta's overall AI computing capacity to 14 gigawatts next year.

Why it matters: This is a concrete production date (not a roadmap slide) for Meta's push to reduce Nvidia dependency, arriving the same week Zuckerberg floated renting out Meta's compute externally — two sides of the same capacity story.

Source: TechCrunch: Meta's new AI chips will begin production in September

🤖 AI Agents & Autonomy

See Tools & Releases above: ChatGPT Work, Claude Cowork's mobile expansion, and AlphaEvolve's general availability are today's three concrete agentic releases — spanning enterprise task delegation (OpenAI, Anthropic) and autonomous algorithm optimization (Google).

🔒 Safety, Alignment & Ethics

Meta's Muse Image Lets Users Generate AI Content From Other People's Photos by Default

Meta's new Muse Image model, live on Instagram Stories (US) and WhatsApp (select countries) since July 9, lets users tag another person's public Instagram profile with "@" to generate new AI images drawn from that person's photos. People are notified when their media is used for remixes, sequences, stickers, or templates — but not when it's used to generate new AI content via Muse Image.

Why it matters: This is a live, shipped consent gap — not a hypothetical — affecting anyone with a public Instagram profile, regardless of whether they use Meta AI themselves.

Source: Washington Times: Meta AI image generation model lets users work off other people's Instagram media

📊 Numbers & Signals

$1.2 trillion — Anthropic's implied secondary-market valuation, today
$908 billion — OpenAI's implied secondary-market valuation, same platforms
550% — Anthropic's valuation increase over the past year
$65 million — Ollama's new Series B (Theory Ventures); $88M total raised
8.9 million — Developers on Ollama; 85% of the Fortune 500 use it
$1.25 / $4.25 — Meta Muse Spark 1.1 API pricing per million input/output tokens
14 gigawatts — Meta's targeted AI computing capacity next year
80% — Supply chain model improvement BASF reported using AlphaEvolve

🧠 Worth Thinking About

Three labs made competing moves on the exact same axis today: OpenAI shipped an autonomous work agent, Anthropic put its work agent on phones, and Meta undercut both labs' API pricing by roughly 4x while pushing its own agentic model. None of this is coincidence — it's convergence. The entire frontier-lab product roadmap for mid-2026 has quietly become "ship an agent that can be handed a goal and left alone for hours," and the valuation story (Anthropic passing OpenAI by $300B+ on paper) suggests investors are pricing that race, not the underlying chat models, as the thing that actually matters now. Worth sitting with: none of these agent products have a public track record yet longer than a few days.

🏛️ Government & Regulation

Nothing genuinely new in the last 24 hours. The Illinois AI Safety Measures Act (signed July 6) and the EU AI Act Digital Omnibus (still awaiting Official Journal publication) are both already covered in prior briefings and unchanged as of today.

🔭 Frontier Lab Dispatch

OpenAI — July 9: Shipped ChatGPT Work, an autonomous task-delegation agent built on GPT-5.6, for Pro/Enterprise/Edu plans.

Anthropic — July 9: Expanded Claude Cowork to web and mobile (iOS/Android); separately, Anthropic's secondary-market valuation hit $1.2T, passing OpenAI's $908B.

Google DeepMind / Google Cloud — July 9: AlphaEvolve, the Gemini-powered algorithm-optimization agent, reached general availability for all Google Cloud customers.

Meta AI — July 9: Launched Muse Spark 1.1 (Meta's first paid model API) and Muse Image (Instagram/WhatsApp); confirmed September production start for its in-house Iris AI chip.

🔗 Quick Links

Tier 1 — Frontier AI Labs

Tier 3 — Tech & AI News Media