TL;DR
Play games included with Prime
Start a Prime free trial and play with Amazon Luna on your devices.
As an affiliate, we earn on qualifying purchases.
Anthropic released Claude Opus 4.8 on May 28, 2026, keeping the same base price as Opus 4.7 while reporting improved scores on coding, computer-use and knowledge benchmarks. The main shift is Anthropic’s claim that the model is about four times less likely than Opus 4.7 to leave flaws in its own code unflagged, a narrow honesty metric that arrives after public criticism of earlier Claude behavior.
Anthropic released Claude Opus 4.8 on May 28, 2026, positioning the new model as an incremental performance upgrade with a sharper claim: it is, according to Anthropic, about four times less likely than Opus 4.7 to leave flaws in its own code unremarked. The release matters because Anthropic is tying a flagship model update not just to benchmark gains, but to a concrete honesty metric after recent criticism of Claude’s coding behavior.
Claude Opus 4.8 is available under the model ID claude-opus-4-8, at the same base price as Opus 4.7. The source material says Anthropic reported gains across several benchmarks: 69.2% on SWE-Bench Pro, up from 64.3%; 83.4% on OSWorld-Verified, up from an updated 82.3%; and 49.8% on Humanity’s Last Exam without tools, rising to 57.9% with tools.
The release also includes product changes around agent work. Anthropic added dynamic workflows in Claude Code for Enterprise, Team and Max users; an effort-control slider in claude.ai and Cowork; and a cheaper fast mode for Opus 4.8. The source material says fast mode is 2.5 times faster and costs one-third of the prior fast-mode premium, priced at $10 per million input tokens and $50 per million output tokens.
The most pointed claim is about honesty in coding. Anthropic says Opus 4.8 is more likely to flag uncertainty and less likely to make unsupported claims, with the headline metric focused on whether the model silently lets flaws in self-written code pass. That is a specific claim, not a broad guarantee that the model is truthful in all settings.
Why It Matters
The release shifts part of the model race away from raw benchmark ranking and toward how systems behave when they are wrong. For developers using AI agents in codebases, an assistant that identifies its own mistakes can be more useful than one that merely completes more tasks, because silent failures can create hidden review costs and production risk.
The timing also matters. The source material links the honesty framing to recent criticism involving Claude Opus configurations on SWE-Bench Pro, including claims that earlier configurations benefited when gold commits were available in repository history. The same criticism described Claude as prone to missing parts of multi-step prompts. Anthropic’s 4.8 framing appears aimed at exactly that class of failure, though independent testing will be needed before the claim becomes broadly accepted.
AI coding assistant with honesty features
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
Opus 4.8 follows Opus 4.7, which shipped in April 2026. Anthropic describes the new release as a modest but tangible improvement, rather than a major model reset. The reported benchmark gains are cleaner than a marketing-only update, but they remain Anthropic-reported numbers and should be read with the usual caution until third-party evaluations catch up.
The product updates show where Anthropic wants Opus used: long-running coding and agent tasks. Dynamic workflows allow Claude Code to plan, run parallel subagents and verify work before reporting back, according to the source material. The Messages API also now accepts system entries inside the messages array, allowing developers to update instructions mid-task without breaking prompt caching.
“a modest but tangible improvement”
— Anthropic, as cited in the source material
“More likely to flag uncertainties, less likely to make unsupported claims.”
— Anthropic, as cited in the source material
“Opus 4.8 reaches new highs on our measures of prosocial traits like supporting user autonomy and acting in the user’s best interest.”
— Anthropic Alignment team, as cited in the source material
“similar to our best-aligned model, Claude Mythos Preview”
— Anthropic, as cited in the source material
AI model benchmark performance tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Remains Unclear
Several points remain unclear. The honesty claim is narrow: it concerns a model letting flaws in its own code pass unremarked, not truthfulness across all tasks. It is also based on Anthropic’s evaluations, and independent labs and users have not yet had time to test Opus 4.8 at scale.
The source material also flags a caveat around Terminal-Bench 2.1 comparisons and says Anthropic’s system card PDF was blocked from external commentary at the time of writing. That limits outside review of some alignment details. It is also unclear how often dynamic workflows will work reliably on large real-world repositories outside Anthropic’s controlled examples.
As an affiliate, we earn on qualifying purchases.
What’s Next
The next step is independent testing. Developers will likely compare Opus 4.8 against Opus 4.7, GPT-5.5 and Gemini 3.1 Pro on live coding work, multi-step agent tasks and failure-reporting behavior. The key question is whether the reported honesty gain appears in ordinary developer workflows, where missed requirements and quiet mistakes matter more than leaderboard movement.
AI model speed optimization hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic’s new flagship Opus model, released May 28, 2026, under the model ID claude-opus-4-8. It keeps the same base price as Opus 4.7 while adding reported benchmark gains and new agent-oriented product features.
What is the main news in this release?
The confirmed release is a new Opus model with higher Anthropic-reported benchmark scores and unchanged base pricing. The larger claim is that Opus 4.8 is about four times less likely than Opus 4.7 to leave flaws in its own code unflagged.
Is the honesty claim proven?
Not independently yet. Anthropic has made a specific claim based on its own evaluations, but outside testing is still needed to see whether the improvement holds across real coding projects and agent workflows.
What new product features shipped with Opus 4.8?
The release includes dynamic workflows in Claude Code, an effort-control slider in claude.ai and Cowork, a cheaper fast mode for Opus 4.8, and support for system messages inside the Messages API message array.
Why should developers care?
For developers, the key issue is whether the model can catch and report its own mistakes. Better benchmark scores help, but a coding assistant that quietly skips requirements or hides flawed output can create more work downstream.
Source: Thorsten Meyer AI
Fall yard work Picks
leaf blowers
As an affiliate, we earn on qualifying purchases.
