Anthropic shipped Claude Opus 4.8 today at 17:18 UTC, just 41 days after Opus 4.7. That is the fastest release cadence in Anthropic's history. The model is available immediately across the Claude API, Amazon Bedrock, Google Vertex AI, and Microsoft Foundry, at the exact same pricing as Opus 4.7. There is no premium for the upgrade. The story most outlets are running is the standard "new flagship model" piece, faster, smarter, slightly higher benchmarks. That coverage is fine, but it is missing what is actually interesting here. Opus 4.8 is the first major frontier release where the headline pitch is not "more capable." It is "more honest." Anthropic is explicitly positioning this model as one that is better at flagging its own uncertainty and admitting when it does not know, and it backed that pitch with real numbers. That is a notable shift in what frontier labs are competing on. Plus, near the bottom of today's announcement, Anthropic confirmed Mythos-class models are coming to all customers "in the coming weeks," directly tying back to the leak story we covered last night.

Here are the verified facts and which ones matter most for your work.

Same price, real upgrade. Standard pricing is unchanged at $5 per million input tokens and $25 per million output tokens. For a frontier model bump, that is unusual. Most upgrade cycles include a price increase to justify the new performance. Anthropic just absorbed the cost improvements internally, which says something about how confident they are in their unit economics with the new SpaceX compute coming online.

Fast Mode is genuinely cheaper. The optional Fast Mode runs at roughly 2.5x the speed of standard Opus 4.8 for $10 input and $50 output per million tokens. That is three times cheaper than Fast Mode was on the previous generation. If you run latency-sensitive workloads, this is the actual margin shift to pay attention to.

The benchmarks that matter. SWE-bench Pro jumped to 69.2% (from 64.3% on Opus 4.7). SWE-bench Verified hit 88.6%. Terminal-Bench 2.1 hit 74.6%. OSWorld-Verified, the agentic computer-use benchmark, hit 83.4%. GDPval-AA, the knowledge-work eval, scored 1890 Elo, 121 points ahead of GPT-5.5 and well ahead of Gemini 3.1 Pro. The honest caveat: GPT-5.5 still wins on agentic terminal coding. Opus 4.8 wins almost everywhere else.

The honesty story is the real story. Anthropic's own measurement: Opus 4.8 is four times less likely than 4.7 to let a code flaw pass without flagging it. On the alignment assessment, it scores at Claude Mythos Preview level for prosocial traits like supporting user autonomy and acting in user best interest. The model is explicitly trained to flag uncertainty instead of confidently guessing. The framing in the official blog post: "A general problem with AI models is that they sometimes jump to conclusions, confidently claiming to have made progress in their work despite the evidence being thin." Opus 4.8 is the answer to that complaint.

Dynamic Workflows is the new product. Available as a research preview in Claude Code, Dynamic Workflows lets Claude orchestrate up to 1,000 parallel subagents through a JavaScript script that runs in the background. Your session stays responsive while the agents work. This is Anthropic's answer to long-horizon agentic tasks like full codebase migrations. Opus 4.8 became the only model to complete all cases end-to-end on the Super-Agent benchmark.

Effort control. A new slider on claude.ai lets users dial up or down how much computational effort Claude puts into a response. "xhigh" and "max" trade speed for depth. Lower settings burn through rate limits more slowly.

Rate limits reset. Anthropic reset all weekly and hourly rate limits at launch, a nice perk that signals confidence in the new compute capacity.

For an audience that has actually deployed AI in production, this is the change that will affect daily work the most.

The dirty secret of every frontier model right now is that they confidently BS. Ask any model whether it finished a task and it will say yes. Ask whether the code passes the tests and it will say yes. Whether the answer is actually right is another matter. The cost of that confident bullshit, in enterprise contexts, is the entire reason humans still review AI output line by line.

Opus 4.8 is the first frontier model where the lab explicitly built the training pipeline around fixing that. A 4x reduction in unreported code flaws is not a marketing number, it is a workflow change. If the model reliably flags "I am not sure this works, please verify" instead of saying "done," the review burden drops meaningfully. The downstream effect is that you can trust longer chains of autonomous work, which is exactly what Dynamic Workflows is designed for.

This also explains why Anthropic is winning enterprise contracts at the rate it is. Goldman, KPMG, Deloitte, PwC, JPMorgan, BlackRock, the customers who actually pay seven and eight-figure annual contracts, do not care about the difference between 87% and 89% on a coding benchmark. They care about whether the model will tell their compliance officer when it is unsure. Anthropic is the only lab making that the central pitch. The valuation premium is starting to make sense.

Shipped today alongside Opus 4.8 as a research preview, Dynamic Workflows lets Claude orchestrate up to 1,000 parallel subagents through a JavaScript script that runs in the background. You give Claude a complex multi-step task, it writes the script, the runtime dispatches the agents, they refute and verify each other's work, and your session stays responsive while they grind through it.

The benchmark consequence: Opus 4.8 became the only model to complete all cases end-to-end on the Super-Agent benchmark. Anthropic also called out a full Bun port case study where Dynamic Workflows handled what would have been a multi-week migration overnight.

For working devs, this is the "fire and forget" agent product the Claude Code stack has been missing. Full codebase migrations, large refactors, multi-repo audits, all become single overnight jobs instead of weeks of babysitting. Available now in Claude Code for Pro, Max, Teams, and Enterprise plans.

One related kicker: Anthropic confirmed today that Mythos-class models reach all customers "in the coming weeks." That ties straight back to last night's send. The Claude Code stack you have in a month is going to look meaningfully different from today.

For working developers, Opus 4.8 is a clear upgrade for any task involving code review, refactoring, or complex agentic workflows. The honesty gains matter most on tasks where you would otherwise have to babysit the output. If you are still on Opus 4.7 in Claude Code, switching to 4.8 is free and the cost of switching back is zero, so the decision is trivial.

For teams running production AI workloads, Fast Mode at $10/$50 is the budget unlock. If you have been priced out of frontier-quality inference on high-volume tasks, this is the first time the math actually works without dropping to a cheaper model. Worth running a real cost comparison this week.

For anyone watching the frontier race, today is the cleanest signal yet that Anthropic believes the next axis of competition is not raw capability, it is reliability and honesty. Every other lab is still optimizing for benchmark wins. Anthropic is optimizing for the model that gets trusted with real money. That is the same bet that took them from $380 billion to $900 billion in three months.

Claude Opus 4.8 shipped today, 41 days after 4.7, the fastest release cadence in Anthropic's history. Same price ($5 input / $25 output per million tokens) with Fast Mode now three times cheaper than the prior generation at $10/$50. SWE-bench Pro hit 69.2%, GDPval-AA hit 1890 Elo (121 points ahead of GPT-5.5), and Opus 4.8 is four times less likely to let a code flaw pass unflagged. Dynamic Workflows shipped alongside it, letting Claude Code orchestrate up to 1,000 parallel subagents on long-running tasks. The strategic pitch is honesty, not raw capability, which is the metric enterprise buyers actually price. And Anthropic confirmed Mythos-class models will reach all customers "in the coming weeks," giving us a real timeline for the leak story we covered last night. The version number says 4.8. The actual story is that Anthropic is now competing on the one axis nobody else is.

Claude Opus 4.8 👉 Here

Claude Code: Dynamic Workflows 👉 Here

Stay building. 🤖