Claude Opus 4.8 Is Here: Effort Control, Dynamic Workflows, and a More Honest Model

On May 28, 2026, Anthropic released Claude Opus 4.8, the latest upgrade to its flagship model. It comes just six weeks after Opus 4.7 and costs the same as the prior model. This update brings a broad set of refinements across coding, reasoning, honesty, and—maybe most interesting for everyday users—how much control you get over the model's behavior.

Here's what changed and why it matters.

Better coding and reasoning, measured

Opus 4.8 scored meaningful gains on the benchmarks Anthropic tracks. Agentic coding jumped from 64.3% to 69.2%, and multidisciplinary reasoning with tools rose from 54.7% to 57.9%. On SWE-Bench Pro, the model hit 69.2%, beating competitors like GPT-5.5 and Gemini 3.1 Pro on most coding and reasoning tasks—though OpenAI's model reportedly still leads on agentic terminal coding specifically.

The more practical improvement is reliability. Anthropic's internal evaluations found Opus 4.8 is about four times less likely than 4.7 to overlook flaws in its own generated code. Early enterprise testers confirmed this.

Effort control: the real product shift

A new control now sits next to the model selector in Claude.ai and Cowork (and as effort levels in Claude Code), letting you decide how hard Claude works on a given response. Higher effort means the model spends more tokens to produce better results; lower effort means faster, cheaper answers for simpler work.

It's available on every plan, including the Free tier, and it defaults to high effort. In Claude Code you can dial up to "extra" (xhigh) or "max" for difficult tasks and long-running asynchronous work, with rate limits raised to accommodate the higher token usage. The concensus on how to use it seems to be: use the expensive, high-effort setting where being wrong is costly, and drop down a tier everywhere else.

Faster, cheaper fast mode

Fast mode now runs at up to 2.5x the speed and costs three times less than previous fast-mode pricing. For high-volume workloads, that's a big difference—and it goes with the theme of giving users more ways to control their token use.

Honesty and alignment gains

Anthropic emphasized improvements in alignment and safety. Its Alignment team concluded Opus 4.8 reaches new highs on traits like supporting user autonomy and acting in the user's best interest.

Under the hood

Opus 4.8 supports a 1 million-token context window by default on the Claude API, Amazon Bedrock, and Vertex AI (200k on Microsoft Foundry), 128k max output tokens, and adaptive thinking. The minimum cacheable prompt length dropped to 1,024 tokens. The model is available through the API, GitHub Copilot (Pro+, Business, and Enterprise), and other platforms.

The bigger picture

Anthropic frames Opus 4.8 as a "modest but tangible" improvement. It's an incremental upgrade rather than a major jump in capability. With the recent rumors and leaked information surrounding its Mythos model, Opus 4.8 may be an upgrade that eases users into control of something a little more powerful—before something bigger arrives.

For now, it's a solid upgrade that's free to try at the same price, with the most useful change being the ability to tell Claude how hard to think.

- Jason