Anthropic released Claude Opus 4.8 on May 28, 2026 — yesterday. So this is as fresh as it gets.
If you've been following the AI model releases this year, you'll know Anthropic has been on a tight two-month cadence. Opus 4.6 dropped in February, Opus 4.7 in April, and Opus 4.8 lands right on schedule. Each release has been meaningful — not just marketing increments.
This one is no different. Here's the honest breakdown of what changed, what's new, and whether it's worth your attention.
What Opus 4.8 Actually Is
Claude Opus 4.8 is Anthropic's newest Opus model — a premium AI model built for advanced coding, AI agents, long-context reasoning, and professional knowledge work.
Anthropic is framing the upgrade around reliability rather than raw speed — at a moment when much of the industry is racing toward faster, more autonomous models. The company's wager is that the next thing buyers will pay for is an AI that knows when it might be wrong.
That positioning is deliberate and worth paying attention to. While competitors are chasing benchmark headlines, Anthropic is betting on trust. For developers building production systems, that's actually the right bet.
The Biggest Improvement — Agentic Coding
This is the headline number and it's genuinely significant.
Opus 4.8 leads the pack on agentic coding with 69.2%, compared to 64.3% for Opus 4.7, 58.6% for GPT-5.5, and 54.2% for the next competitor.
That's not a small gap. On the benchmark that matters most for developers building agent-powered workflows, Opus 4.8 is ahead of both GPT-5.5 and Gemini 3.1 Pro by a meaningful margin.
Opus 4.8 also scores 88.6% on SWE-bench Verified and 74.6% on Terminal-Bench 2.1. SWE-bench is the gold standard for real-world software engineering tasks — it tests whether a model can actually fix bugs in real GitHub repositories, not just answer coding questions. 88.6% is a strong number.
Opus 4.8 is roughly four times less likely to leave flaws in its own code unflagged compared to Opus 4.7. That's the reliability angle Anthropic keeps coming back to. An AI that catches its own mistakes before you do is genuinely more useful in production.
Dynamic Workflows — The Feature Worth Paying Attention To
This is the most interesting new capability in this release.
For the hardest tasks, Claude makes a plan, runs hundreds of parallel subagents, and verifies its work before reporting back. Think a migration touching hundreds of files, a codebase audit, a full test suite generation. Tasks that used to require a team now run as a single instruction.
Claude Code has a new Dynamic Workflows feature that allows it to tackle very large-scale problems. The orchestrator-subagent architecture means that instead of one model working through a massive task sequentially, it spins up parallel workstreams that each handle a piece of the problem simultaneously.
For solo developers or small teams with big backlogs, this changes the math on what's possible. A codebase audit that would have taken a week of careful manual work can now be kicked off as a single instruction and reviewed when it's done.
Effort Controls — A Practical Feature That Saves Money
Users on Claude can now select how much thinking effort Claude applies — from Low for faster lower-cost responses, to Max for the hardest problems. Opus 4.8 defaults to High effort for the best balance of quality and experience.
This matters more than it might sound. Running Low effort on simple tasks and Max effort on the hard ones is the discipline that cuts your monthly bill significantly without touching output quality on what matters.
Previously you were paying the same rate regardless of whether you were asking Claude to summarise a paragraph or debug a complex multi-file architecture problem. Now you can match the compute to the task. For heavy API users, this will make a real difference to costs.
Fast Mode — 2.5x Speed, 3x Cheaper Than Before
A faster mode runs at 2.5 times the speed and costs $10 per million input tokens and $50 per million output tokens — three times cheaper than fast mode on previous models.
So you get significantly faster responses at a fraction of what fast mode used to cost. For latency-sensitive applications — real-time coding assistants, interactive tools, anything where the user is waiting — this is a meaningful improvement.
Pricing and Availability
Pricing is unchanged from Opus 4.7 at $5 per million input tokens and $25 per million output tokens. Same cost, meaningfully better model. That's a good deal.
Opus 4.8 is available across consumer, business, developer, and enterprise surfaces. For Claude users it's available on supported Pro, Max, Team, and Enterprise plans. For developers it's available through the Claude API with the model ID claude-opus-4-8. Enterprise teams can also access it through Amazon Web Services, Google Cloud, and Microsoft Foundry.
Claude Opus 4.8 is also now available in GitHub Copilot for Copilot Pro+, Business, and Enterprise users. If you're already using Copilot, you can select it directly from the model picker in VS Code across chat, ask, edit, and agent modes.
The model supports a 1 million token input context window with up to 128K output tokens. For developers working with large codebases, long documents, or complex multi-file projects, that context window is one of the most practically useful things about the Opus line.
Opus 4.8 vs Sonnet — Which Should You Use?
This is the question most developers will have. Here's the honest answer.
Sonnet is faster and cheaper. For most everyday tasks — code completion, quick explanations, standard refactors — Sonnet is the right call. You don't need a sledgehammer for every nail.
Opus 4.8 is for the hard stuff. Complex multi-file architecture decisions. Large-scale code migrations. Tasks where getting it wrong costs you significant time to undo. Agentic workflows where the model needs to reason across many steps without losing the thread.
The effort controls make this easier to navigate now. Start on Low or Medium effort for routine tasks, switch to Opus with Max effort when the problem genuinely needs it. Let the task complexity drive the model choice.
The Honest Take
Opus 4.8 is an incremental step, not a leap. Anthropic isn't claiming otherwise. But the improvements are in exactly the right places — reliability, agentic performance, and cost efficiency for fast mode.
For developers building production AI systems or agent-powered workflows, the Dynamic Workflows feature alone makes this worth upgrading to. The four-times improvement in self-flagging code flaws is the kind of reliability gain that matters when you're shipping real software.
The broader story is Anthropic's consistent cadence. Every two months, a meaningful improvement. No vaporware, no benchmark-only announcements. If you're building on top of Claude, that predictable pace of improvement is itself a feature worth considering.
🛠 Dev Tip of the Week
If you're using Opus 4.8 via the API, experiment with effort controls from day one. Set Low effort as your default for simple queries, and only bump to Max for genuinely complex reasoning tasks. Then compare the output quality and token costs across a week of usage. Most developers will find they can drop to Medium effort for 70% of their tasks without any noticeable drop in quality — and the cost savings add up fast.
If you're already testing it and have an early take, hit reply — would love to hear what you're seeing in the real world.