Anthropic Hits $965B Valuation to Top OpenAI, Drops Claude Opus 4.8

Anthropic announced Thursday it has raised $65 billion in Series H funding at a $965 billion post‑money valuation, according to Reuters. The round, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, more than doubles the company's $380 billion valuation from February and officially makes Anthropic the world's most valuable AI startup.

The funding includes $15 billion of previously committed investments from hyperscalers, including $5 billion from Amazon, plus strategic investments from chipmakers Micron, Samsung, and SK Hynix, Anthropic confirmed. Co‑leads Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ, and XN joined alongside heavyweights like Blackstone, Brookfield, Fidelity, and General Catalyst.

Anthropic CFO Krishna Rao said the funding will "help us serve the historic demand we are experiencing, stay at the research frontier, and bring Claude to more of the places where work happens," according to CNBC. The company's revenue run rate crossed $47 billion earlier this month, up from $30 billion earlier this year and just $10 billion in annual revenue last year.

Anthropic is no longer the scrappy runner‑up. The $965 billion valuation puts it ahead of OpenAI, which closed a $122 billion round in March at an $852 billion post‑money valuation, CNBC reported. The gap is notable: Anthropic is now worth $113 billion more than its chief rival.

The momentum shift is visible in user adoption, too. ChatGPT's share of global AI app downloads fell from 67% in Q2 2025 to 47% in Q2 2026 so far, while Claude jumped from 1% to 14% over the same period, according to SensorTower data reported by Forbes. Claude Code, Anthropic's AI coding assistant, has been a major driver of enterprise adoption and revenue growth.

Anthropic has also locked in massive compute capacity. The company signed agreements with Amazon for five gigawatts of new capacity, with Google and Broadcom for five gigawatts of advanced TPU capacity, and with SpaceX for GPU access at its Colossus data centers, Anthropic noted. Claude is now the first frontier model available on all three major cloud platforms: AWS, Google Cloud, and Microsoft Azure.

Alongside the funding news, Anthropic released Claude Opus 4.8, calling it the company's "most honest" model yet. Early testers reported the model is four times less likely than its predecessor to allow flaws in code it has written to pass unremarked, according to Anthropic's system card. The company said Opus 4.8 is "more likely to flag uncertainties about its work and less likely to make unsupported claims."

On benchmarks, Opus 4.8 posted a 69.2% on SWE‑Bench Pro, the coding agent benchmark — up from Opus 4.7's 64.3% and OpenAI's GPT‑5.5 at 58.6%. On GDPval, an OpenAI‑created benchmark measuring economically viable work, Opus 4.8 scored 1,890 compared to 1,753 for Opus 4.7 and 1,769 for GPT‑5.5, Inc. Magazine reported.

Harvey, the legal AI company, recorded its highest‑ever score on its internal legal agent benchmark using Opus 4.8. Niko Grupen, Harvey's head of applied research, said the model delivers "the kind of accuracy lift that translates directly into how much real attorney work our customers can hand off with confidence,".⁵ Pricing remains unchanged from Opus 4.7 at $5 per million input tokens and $25 per million output tokens, with fast mode now three times cheaper than before at $10/$50 per million tokens. Grupen shared the results in.⁵

Opus 4.8 ships with several features aimed squarely at developers. Dynamic workflows, available in research preview for Claude Code, lets the model plan large tasks and spawn hundreds of parallel subagents in a single session — then verify outputs before reporting back. Anthropic says Claude Code with Opus 4.8 can now handle "codebase‑scale migrations across hundreds of thousands of lines of code from kickoff to merge," according to Anthropic.

Effort control is a new dropdown that lets users choose how much compute Claude dedicates to a task. Higher effort means deeper thinking and better results; lower effort means faster responses and longer rate limit endurance. Users can select from default, extra, and max levels.

For API developers, the Messages API now accepts system entries inside the messages array, allowing mid‑task instruction updates without breaking prompt caching. This is particularly useful for updating permissions, token budgets, or environment context as an agent runs. Cursor, the AI code editor, reported that Opus 4.8 "exceeds prior Opus models across every effort level" on CursorBench, with more efficient tool calling using fewer steps for the same intelligence,.⁵

Hours after Anthropic's announcement, Elon Musk posted on X that the SpaceX‑Anthropic compute deal — portrayed in SpaceX's IPO filing as a $1.25 billion per month contract through May 2029 — is actually "a 180 day lease with 90 day notice mutual cancellation thereafter," according to the Financial Times. Musk added that the short‑term structure "was our request, not Anthropic's" and that SpaceX might need the capacity back "if compute gets super tight."

A person familiar with the contract confirmed to the FT that the deal has a minimum length of 180 days but is open‑ended beyond that. The discrepancy matters because the $45 billion total value of the deal helped underpin SpaceX's narrative as an AI cloud provider ahead of its IPO. SpaceX's prospectus does disclose a 90‑day termination clause, but does not characterize the deal as a short‑term lease.

The tweet came during SpaceX's SEC‑mandated quiet period following its IPO filing. Harvard Law professor John Coates, a former acting director of the SEC's corporate finance division, told the Financial Times that regulators were unlikely to take action, noting the SEC "shows no apparent interest in enforcing quiet periods." Musk's post was reported by the Financial Times.

Anthropic is preparing to release Mythos‑class models — its most capable AI systems — to all customers "in the coming weeks," the company said. Currently, Claude Mythos Preview is available only to a select group of organizations for cybersecurity work under Project Glasswing. These models require stronger cyber safeguards before general release, and Anthropic says it is making "swift progress" on those,.⁵

On the financial front, all three major AI labs are racing toward public markets. OpenAI is preparing to file its confidential IPO prospectus in the coming days or weeks, targeting a listing as soon as September, CNBC confirmed. SpaceX, which absorbed Musk's xAI earlier this year, filed its prospectus last week. Anthropic is also preparing its IPO "behind the scenes," though timing remains fluid. Apollo and Blackstone are reportedly working on a $36 billion debt deal for Anthropic, according to Reuters.

Anthropic's ascendancy matters for developers in concrete ways. Claude Code is now the dominant AI coding assistant in enterprise environments, and Opus 4.8's dynamic workflows mean single developers can orchestrate work that previously required teams. The effort control feature lets builders trade speed for quality on a per‑task basis — crank it to max for complex refactors, dial it down for quick boilerplate.

The compute deals — with Amazon, Google, and SpaceX — signal that capacity will keep pace with demand, reducing the usage limits and peak‑hour throttling that have frustrated Claude users in recent months. And with all three major AI labs headed for IPOs, 2026 is shaping up as the year AI tooling goes from experimental to industrial‑grade.

The biggest variable is the Musk‑SpaceX compute arrangement. If the deal truly is a 180‑day lease renewable at SpaceX's discretion, Anthropic's infrastructure story has a fragility that its Amazon and Google partnerships may need to compensate for. For now, though, the numbers speak for themselves: $965 billion valuation, $47 billion revenue run rate, and a model that's measurably better at coding and agentic work than anything else on the market.