Anthropic just released Claude Opus 4.8, an upgrade to its most capable model. It builds on Opus 4.7 with meaningful improvements across benchmarks—and it’s available right now at the same price.
But the headline isn’t just a smarter model. Anthropic is also rolling out new features alongside it: effort controls, a major Claude Code upgrade called Dynamic Workflows, and significantly cheaper fast mode pricing.
Here’s what’s actually changed.
1. Better Judgment on Agentic Tasks
The most consistent feedback from early testers? Opus 4.8 makes better decisions when working on its own.
Testers reported that it:
- Asks better questions before diving in
- Catches its own mistakes more reliably
- Pushes back when a plan doesn’t make sense
- Carries tasks through to completion more consistently
That last point matters a lot for teams using Claude in automated workflows. Fewer dropped handoffs and mid-task failures mean less babysitting.
On Anysphere’s CursorBench, Opus 4.8 outperformed previous Opus models across every effort level. On one company’s internal Super-Agent benchmark, it was the only model to complete every case end-to-end—beating both prior Opus versions and GPT-5.5 at the same cost.
2. A More Honest Model
One of the less flashy—but genuinely important—improvements in Opus 4.8 is its honesty.
A common problem with AI models is overconfidence. They’ll claim to have completed work, or ship code with bugs, without flagging the issues. Opus 4.8 is significantly more likely to surface its own uncertainties.
According to Anthropic’s evaluations, Opus 4.8 is around four times less likely than Opus 4.7 to let flaws in its code pass without comment. For professional workflows where accuracy matters—legal, financial, medical—that’s a meaningful improvement.
Anthropic’s alignment team also noted that Opus 4.8 “reaches new highs” on measures of prosocial behavior, including supporting user autonomy and acting in users’ best interests.
3. Dynamic Workflows in Claude Code
This is the biggest new feature for developers and technical teams.
Dynamic Workflows—now in research preview—lets Claude Code tackle problems at a much larger scale. Claude can plan out a task, spin up hundreds of parallel subagents in a single session, and verify outputs before reporting back.
In practice, that means Claude Code with Opus 4.8 can now handle things like:
- Codebase-scale migrations across hundreds of thousands of lines of code
- End-to-end completion from kickoff to merge
- Automated verification against existing test suites
Dynamic Workflows is available for Enterprise, Team, and Max plan users.
4. Effort Controls
Users on claude.ai and Cowork now have a new control alongside the model selector: the ability to choose how much effort Claude puts into a response.
- Higher effort: More thinking, better outputs, uses rate limits faster
- Lower effort: Faster responses, slower rate limit usage
This is available on all plans. For most everyday tasks, the default setting will be fine. For complex research, long documents, or difficult coding problems, bumping up the effort level could make a real difference.
Developers also get a new API feature: system entries can now be passed inside the messages array, letting teams update Claude’s instructions mid-task without breaking prompt caches.
5. Fast Mode Is Now 3× Cheaper
Fast mode—which runs Opus 4.8 at 2.5× the standard speed—is now three times cheaper than it was for previous models.
Pricing for standard usage is unchanged from Opus 4.7:
- Input: $5 per million tokens
- Output: $25 per million tokens
Fast mode pricing:
- Input: $10 per million tokens
- Output: $50 per million tokens
For teams running high-volume workflows where speed matters, this is a meaningful cost reduction.
6. Strong Results in Legal, Finance, and Browser Tasks
Testers across specialized industries saw standout performance:
- Legal: Opus 4.8 achieved the highest score recorded on one firm’s Legal Agent Benchmark—and is the first model to break 10% on an all-pass standard for substantive legal work
- Finance: Testers reported higher quality analysis, denser outputs, and—notably—a stronger tendency to proactively flag issues with inputs and outputs that other models missed
- Browser agents: Opus 4.8 scored 84% on Online-Mind2Web, a significant jump over both Opus 4.7 and GPT-5.5 for computer-use and browser-agent tasks
What’s Coming Next
Anthropic signaled that Opus 4.8 is a stepping stone, not a ceiling.
Two things are in the pipeline: models that deliver Opus-level capabilities at lower cost, and a new, higher-intelligence model class tied to Project Glasswing. A limited number of organizations are already using Claude Mythos Preview for cybersecurity work. Anthropic says it expects to bring Mythos-class models to all customers in the coming weeks.
Why This Release Matters
Claude Opus 4.8 isn’t a dramatic reinvention—but it’s a focused, practical upgrade that hits the areas that matter most for professional use.
The combination of better judgment, improved honesty, Dynamic Workflows, and cheaper fast mode makes this a meaningful update for teams already building on Claude—and a strong case for anyone who hasn’t evaluated it recently to take another look.
Claude Opus 4.8 is available today. Developers can access it via the Claude API using claude-opus-4-8.
Leave a Reply