Claude Opus 4.8 Released—Faster, More Honest; Anthropic’s $65B Funding Surpasses OpenAI

Opus 4.8: Smarter and More Honest

Claude Opus 4.8 arrived only six weeks after Opus 4.7, accelerating Anthropic’s release cadence. Pricing stays at $5 per million input tokens and $25 per million output tokens, with a 1 M‑token context window.

In seven benchmark tests Opus 4.8 wins six. SWE‑Bench Pro improves by 4.9 percentage points (64.3 % → 69.2 %), the metric engineers notice most. Terminal‑Bench 2.1 is the only loss, scoring 74.6 % versus GPT‑5.5’s 78.2 %.

Anthropic’s internal evaluation shows Opus 4.8’s likelihood of silently releasing code defects is only a quarter of the previous generation, and it more often reports uncertainty instead of claiming completion.

Effort level defaults to “high” (down from “xhigh”), consuming similar token counts but achieving higher scores. Users can select “extra” (formerly “xhigh”) or “max” for tougher problems. The “Fast” mode runs about 2.5 × faster than standard while costing one‑third of the previous Fast mode.

Claude Code adds Dynamic Workflows (research preview). Engineers can split a large task into hundreds of parallel sub‑agents that verify, refute, and iterate before converging on a final answer. Using this, Claude Code can migrate libraries spanning hundreds of thousands of lines of code, using existing test suites as acceptance criteria.

The Messages API now accepts a “system” entry within the message array, allowing developers to update instructions mid‑agent run without breaking prompt caching.

Mythos: A Model One Tier Above Opus

Anthropic plans a cheaper, same‑capability model and a more capable new tier called Claude Mythos (code‑named “Copybara”), positioned above Haiku, Sonnet, and Opus. Mythos Preview was announced on April 7 alongside Project Glasswing.

Project Glasswing is a joint security initiative involving AWS, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks, aiming to give defenders a lasting edge in AI‑driven cyber‑warfare.

In the CyberGym security benchmark Mythos Preview scores 83.1 % versus Opus 4.6’s 66.6 %, a 17‑point gap. Access is limited to Glasswing partners for defensive work, with a broader release to all customers expected in coming weeks.

Valuation Surpasses OpenAI

Anthropic closed a $65 billion Series H round, valuing the company at $96.5 billion—higher than OpenAI’s $85.2 billion post‑money valuation from March. Lead investors include Altimeter Capital, Dragoneer, Greenoaks, and Sequoia, each contributing over $2 billion; other backers are D.E. Shaw, BlackRock, DST Global, and chip makers Micron, Samsung, and SK Hynix.

Both Anthropic and OpenAI are expected to file confidential IPO paperwork within weeks to months, making the race to go public a pivotal next step.

Overall, Opus 4.8 delivers a regular product upgrade with unchanged pricing, higher performance, improved honesty, and dynamic workflows for larger tasks, while Mythos showcases the next generation of capability and security, and the massive funding round validates Anthropic’s multi‑track strategy.