Anthropic’s Claude Opus 4.8, the latest iteration of its most advanced publicly available large language model, landed Thursday along with a financial milestone.
The San Francisco-based startup officially surpassed rival OpenAI in valuation, climbing to $965 billion compared to OpenAI’s $852 billion valuation recorded in March. The shift positions Anthropic as the most valuable artificial intelligence (AI) startup in the world.
The release of Opus 4.8 comes just 41 days after the launch of Opus 4.7, an unusually aggressive upgrade cycle for Anthropic, which typically spaces major model updates by three to seven months. Industry analysts suggest the swift turnaround was prompted by a lukewarm reception to Opus 4.7, alongside mounting competitive pressure from recent updates to OpenAI’s Codex and Google’s Gemini Flash.
While Opus 4.8 posts modest improvements on standard industry benchmarks, Anthropic is heavily leaning into the model’s enhanced reliability, safety, and transparency. The company’s Alignment team reported that the model achieves “new highs” in prosocial traits, exhibiting substantially lower rates of deceptive or misaligned behavior compared to its predecessor.
According to Anthropic, early testers found the model to be four times less likely than Opus 4.7 to allow flaws in its written code to pass unremarked.
“One of the most prominent improvements in Opus 4.8 is its honesty,” the company said in a blog post. The model, it said, is now significantly more likely to flag uncertainties and push back against unsupported claims.
Enterprise clients are already noting the difference. Investment management firm Bridgewater Associates reported that Opus 4.8 proactively flags issues within the inputs and outputs of complex analyses — a critical safeguard that other market models routinely miss. Tom Pritchard, a staff engineer at Shopify Inc., echoed the sentiment, praising the model’s improved judgment and its ability to catch its own mistakes during complex programming tasks.
Alongside the model, Anthropic rolled out a research preview of Dynamic Workflows, a feature designed to manage massive, large-scale engineering tasks. The system allows Opus 4.8 to autonomously plan work, deploy and manage hundreds of parallel subagents within a single session, and adapt its priorities on the fly based on real-time findings.
Anthropic claims that when paired with Claude Code, the system can execute codebase-scale migrations across hundreds of thousands of lines of code, relying on the subagents to vet and verify data before reporting back to human supervisors.
Additionally, Anthropic is introducing an Effort Control feature to Claude.ai and Cowork. Users can now manually adjust the amount of computational power the model uses for high effort, which prompts the model to think more deeply and frequently for complex engineering, and low effort, which delivers faster response times while conserving user rate limits.
“Frontier models now ship on a monthly cadence, turning the model layer into a moving dependency that development teams must continuously re-validate,” said Mitch Ashley, vice president and practice lead for Software Lifecycle Engineering and AI-Native Software Engineering at The Futurum Group. “Each release shifts the behavior agentic and coding workflows were tuned around. Gains in code-generation reliability and agent judgment only deliver value once teams verify a new model against their own pipelines. Continuous validation has become a standing operating cost of building on frontier models.”
Claude Opus 4.8 is available immediately across the Claude platform and API. In a move to appease users, Anthropic has kept standard token pricing identical to Opus 4.7, holding steady at $5 per million input tokens and $25 per million output tokens. However, the company revealed that its Fast Mode — which operates at 2.5 times standard speed — will now be three times cheaper than it was for previous iterations.
Looking ahead, Anthropic hinted that the preview period for its highly anticipated, next-generation Mythos class of models may soon draw to a close. The advanced model was held back last month due to cybersecurity concerns, but the company stated it is making “swift progress” on necessary safeguards and expects a wider rollout in the coming weeks.
