Anthropic is having a busy week. On Thursday, the San Francisco-based AI lab introduced it had raised $65 billion in a new funding spherical that valued the corporate at $965 billion, making it essentially the most priceless AI startup in the world.
On the identical day, it additionally shipped Claude Opus 4.8, a new AI mannequin that Anthropic says is much less more likely to deceive customers or cooperate with misuse than its predecessors. In the identical announcement, Anthropic mentioned it expects to convey its Mythos-class fashions to all prospects “in the coming weeks.”
Mythos, the existence of which was first reported by Fortune in March, is notable for its coding and cyber capabilities, together with the flexibility to search out vulnerabilities in present software program and chain these vulnerabilities collectively to execute subtle cyber assaults. The mannequin has raised issues that it might permit hackers to simply compromise the networks and software program that management important infrastructure, from monetary techniques to electrical grids, and Anthropic has thus far allowed solely a choose group of customers to entry the mannequin in order to search out vulnerabilities in their very own software program and patch them. It mentioned it will not launch the mannequin extra broadly till it had developed safeguards to make it harder for hackers to take advantage of its capabilities for malicious functions.
The spate of reports comes as Anthropic races towards one of the vital anticipated IPOs of the 12 months, with the corporate reportedly quietly engaged on public market preparations even because it continues to lift at record-breaking personal valuations.
In February, Anthropic was valued at $380 billion after closing its Collection G. Since then, the corporate has gained vital momentum: it has seen a surge in client customers, widespread adoption of its coding device, Claude Code, by companies, and the debut of its highly effective Mythos mannequin in a restricted preview. The corporate’s enterprise gross sales progress has been significantly sturdy. By Could, the corporate mentioned its annualized run-rate income had crossed $47 billion, up from simply $10 billion on the finish of 2025—a fee of progress that is unprecedented for a firm of this measurement.
Anthropic’s most up-to-date funding spherical, its Collection H, was led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with Capital Group, Coatue, D1 Capital Companions, GIC, ICONIQ, and XN co-leading. Of the $65 billion complete, $15 billion represents beforehand dedicated funding from hyperscalers, together with $5 billion from Amazon.
The numbers are spectacular even by the skyrocketing valuations of the present AI growth—it additionally leapfrogs rival OpenAI’s most up-to-date valuation of $852 billion, set when the corporate closed its personal record funding spherical in March.
Anthropic’s CFO Krishna Rao mentioned the brand new funding would assist the corporate serve the brand new “historic demand” the corporate was experiencing. The corporate has beforehand acknowledged it was struggling to serve the brand new demand for Claude, telling Fortune that its computing infrastructure had been stretched to satisfy the unprecedented progress, significantly at peak hours. The corporate has signed new compute offers since then, together with a reported $1.25 billion a month settlement with Elon Musk’s xAI working by way of Could 2029.
Opus 4.8
On Thursday, Anthropic additionally launched a new frontier mannequin from its Opus tier, simply over a month because it launched Opus 4.7. Anthropic’s personal evaluations discovered that Opus 4.8 exhibits charges of misaligned conduct (together with deception and cooperation with misuse) which are “considerably decrease than Opus 4.7” and corresponding to Claude Mythos Preview, the corporate’s most superior mannequin.
The corporate says 4.8 is round 4 instances much less doubtless than 4.7 to let flaws in its personal code move unremarked—presumably a response to current customers complaints about sloppy code generated by Claude Code—and that it “reaches new highs on our measures of prosocial traits like supporting person autonomy and performing in the person’s greatest curiosity.”
On benchmark evaluations from Cursor, Synthetic Evaluation, and Anthropic’s personal Tremendous-Agent benchmark the mannequin outperforms its predecessor, 4.7, on agentic coding and long-horizon duties.
Nonetheless, the reception from customers and different evaluators has been combined.
Andon Labs discovered that 4.8 efficiency was worse than 4.7 on sure exams of enterprise process and on so-called “adversarial benchmarks” that are designed to probe the mannequin’s weaknesses fairly than its strengths, characterising the mannequin as extra cautious and “frightened of getting caught.” Some customers on-line have additionally complained concerning the mannequin being restrained and combative, refusing to conform with even simple and benign requests.
Pricing for the brand new mannequin is the identical $5 per million enter tokens and $25 per million output tokens as for its predecessor 4.7. However “quick mode”—a totally different API configuration that generates tokens from Anthropic’s Claude Opus fashions a lot sooner, however which additionally price customers extra per token—is now 3 times cheaper than it was for 4.7, which can soften the general price for high-volume customers.
Source link
#Anthropic #leapfrogs #OpenAI #record #billion #valuation #Mythos #coming #weeks #Fortune
