Anthropic is having a busy week. On Thursday, the San Francisco-based AI lab announced it had raised $65 billion in a new funding round that valued the company at $965 billion, making it the most valuable AI startup in the world.
On the same day, it also shipped Claude Opus 4.8, a new AI model that Anthropic says is less likely to deceive users or cooperate with misuse than its predecessors. In the same announcement, Anthropic said it expects to bring its Mythos-class models to all customers “in the coming weeks.”
Mythos, the existence of which was first reported by Fortune in March, is notable for its coding and cyber capabilities, including the ability to find vulnerabilities in existing software and chain these vulnerabilities together to execute sophisticated cyber attacks. The model has raised concerns that it could allow hackers to easily compromise the networks and software that control critical infrastructure, from financial systems to electric grids, and Anthropic has so far allowed only a select group of users to access the model in order to find vulnerabilities in their own software and patch them. It said it would not release the model more widely until it had developed safeguards to make it more difficult for hackers to exploit its capabilities for malicious purposes.
The spate of news comes as Anthropic races toward one of the most anticipated IPOs of the year, with the company reportedly quietly working on public market preparations even as it continues to raise at record-breaking private valuations.
In February, Anthropic was valued at $380 billion after closing its Series G. Since then, the company has gained significant momentum: it has seen a surge in consumer users, widespread adoption of its coding tool, Claude Code, by businesses, and the debut of its powerful Mythos model in a limited preview. The company’s enterprise sales growth has been particularly strong. By May, the company said its annualized run-rate revenue had crossed $47 billion, up from just $10 billion at the end of 2025—a rate of growth that is unprecedented for a company of this size.
Anthropic’s most recent funding round, its Series H, was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ, and XN co-leading. Of the $65 billion total, $15 billion represents previously committed investment from hyperscalers, including $5 billion from Amazon.
The numbers are impressive even by the skyrocketing valuations of the current AI boom—it also leapfrogs rival OpenAI’s most recent valuation of $852 billion, set when the company closed its own record funding round in March.
Anthropic’s CFO Krishna Rao said the new funding would help the company serve the new “historic demand” the company was experiencing. The company has previously acknowledged it was struggling to serve the new demand for Claude, telling Fortune that its computing infrastructure had been stretched to meet the unprecedented growth, particularly at peak hours. The company has signed new compute deals since then, including a reported $1.25 billion a month agreement with Elon Musk’s xAI running through May 2029.
On Thursday, Anthropic also released a new frontier model from its Opus tier, just over a month since it released Opus 4.7. Anthropic’s own evaluations found that Opus 4.8 shows rates of misaligned behavior (including deception and cooperation with misuse) that are “substantially lower than Opus 4.7” and comparable to Claude Mythos Preview, the company’s most advanced model.
The company says 4.8 is around four times less likely than 4.7 to let flaws in its own code pass unremarked—possibly a response to recent users complaints about sloppy code generated by Claude Code—and that it “reaches new highs on our measures of prosocial traits like supporting user autonomy and acting in the user’s best interest.”
On benchmark evaluations from Cursor, Artificial Analysis, and Anthropic’s own Super-Agent benchmark the model outperforms its predecessor, 4.7, on agentic coding and long-horizon tasks.
However, the reception from users and other evaluators has been mixed.
Andon Labs found that 4.8 performance was worse than 4.7 on certain tests of business task and on so-called “adversarial benchmarks” which are designed to probe the model’s weaknesses rather than its strengths, characterising the model as more cautious and “scared of getting caught.” Some users online have also complained about the model being restrained and combative, refusing to comply with even straightforward and benign requests.
Pricing for the new model is the same $5 per million input tokens and $25 per million output tokens as for its predecessor 4.7. But “fast mode”—a different API configuration that generates tokens from Anthropic’s Claude Opus models much faster, but which also cost users more per token—is now three times cheaper than it was for 4.7, which may soften the overall cost for high-volume users.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。