Claude Opus 4.8 just launched — and it’s one of the most significant AI model releases of the year. Announced on May 28, 2026, Anthropic’s latest upgrade to its flagship model brings stronger coding performance, smarter agentic behavior, major honesty improvements, and new features that make it more useful for real-world, long-running tasks.
If you use Claude for development, writing, research, or complex professional work, here’s everything you need to know about what changed and why it matters.
What Is Claude Opus 4.8?
Claude Opus 4.8 is the newest version of Anthropic’s most capable AI model. It’s a direct upgrade to Opus 4.7 (released April 16, 2026), arriving less than two months later — a sign of Anthropic’s accelerating release cadence.
Anthropic describes it as a “more effective collaborator” — a model with sharper judgment, more honesty about its progress, and the ability to work independently for longer periods than its predecessors.
The model is available today across all platforms at the same price as Opus 4.7.
What’s New in Claude Opus 4.8?
Stronger Benchmark Performance
Claude Opus 4.8 improves across every major benchmark category:
- SWE-Bench Pro (agentic coding): 69.2% — up from 64.3% on Opus 4.7
- Terminal-Bench 2.1 (command-line tasks): 74.2% — an 8.4% jump over 4.7
- Multidisciplinary reasoning with tools: 57.9% — up from 54.7%
- Online-Mind2Web (computer use and browser agent): 84%
According to Anthropic, Opus 4.8 outperforms GPT-5.5 and Gemini 3.1 Pro on several key benchmarks, including agentic coding, reasoning, and financial analysis — though GPT-5.5 still leads on terminal-coding specifically.
Better Honesty and Judgment
This is one of the most meaningful upgrades in Opus 4.8, and it’s not about raw performance numbers.
A persistent problem with AI models is overconfidence — jumping to conclusions and claiming progress even when the evidence is thin. Opus 4.8 directly addresses this.
Early testers report that Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims. In concrete terms, the model is around four times less likely than its predecessor to allow flaws in code to pass without flagging them.
One tester described it this way: Opus 4.8 “proactively flags issues with the inputs and outputs of an analysis, something other models routinely missed and left to the users to catch.”
Improved Alignment and Safety
Anthropic runs a detailed alignment assessment on every model before release. For Opus 4.8, the results are notably strong.
The assessment found that Opus 4.8 “reaches new highs on measures of prosocial traits like supporting user autonomy and acting in the user’s best interest.” Rates of misaligned behavior — including deception and cooperation with misuse — are substantially lower than Opus 4.7, and on par with Claude Mythos Preview, Anthropic’s most aligned model.
Fast Mode: 2.5x Speed, 3x Cheaper
Fast mode is back and significantly improved. In fast mode, Opus 4.8 can work at 2.5 times the speed of the standard model. The cost of fast mode is now three times cheaper than it was for previous versions.
This makes fast mode genuinely viable for developers and businesses who need speed without paying a premium for it.
New Features Launching With Opus 4.8
Dynamic Workflows (Research Preview)
This is the headline new feature, and it’s a big one for developers.
Dynamic Workflows allows Claude to take on very large-scale problems in Claude Code. Instead of working through a task linearly, Claude can now:
- Plan the full scope of a project
- Spin up hundreds of parallel subagents in a single session
- Verify outputs before reporting back to the user
In practice, this means Claude Code with Opus 4.8 can carry out codebase-scale migrations across hundreds of thousands of lines of code — from kickoff to merge — using the existing test suite as its benchmark.
This is available in research preview for Claude Code Enterprise, Team, and Max plan users.
Effort Control in Claude.ai and Cowork
Users on claude.ai and Cowork now have a new control alongside the model selector: effort level.
- Higher effort — Claude thinks more frequently and more deeply, delivering better responses at the cost of more time and rate limit usage
- High effort (default) — Opus 4.8’s default. The best balance of quality and experience for most tasks
- Lower effort — Claude responds faster and uses rate limits more slowly, suitable for quick tasks
This gives users real control over the tradeoff between speed and thoroughness depending on what they’re working on.
Messages API Mid-Task System Prompts
For developers, the Messages API now accepts system entries inside the messages array. This means developers can update Claude’s instructions mid-task — useful for long-running agentic workflows where context or requirements shift during execution.
What Real Teams Are Saying
Opus 4.8 has already been tested by a wide range of companies. Here’s what some of them reported.
Cursor (AI code editor): On CursorBench, Opus 4.8 exceeds prior Opus models across every effort level. Tool calling is meaningfully more efficient, using fewer steps for the same intelligence, and it carries end-to-end tasks through.
Legal work: Opus 4.8 delivered the highest score recorded on one Legal Agent Benchmark, and was the first model to break 10% overall on the all-pass standard for complex legal reasoning.
Financial analysis: Teams noted that Opus 4.8 produced richer, more information-dense outputs while finishing faster — and that it proactively flagged issues that other models left for users to catch.
Databricks (enterprise data and AI): Opus 4.8 unlocks a step change in agentic reasoning, tackling deeper, multistep questions faster than prior Opus versions — and at 61% cheaper token cost than Opus 4.7 for multimodal work.
Claude Opus 4.8 vs Opus 4.7: Key Differences
| Feature | Opus 4.7 | Opus 4.8 |
|---|---|---|
| SWE-Bench Pro (agentic coding) | 64.3% | 69.2% |
| Terminal-Bench 2.1 | 65.8% | 74.2% |
| Multidisciplinary reasoning | 54.7% | 57.9% |
| Code flaw flagging | Baseline | 4x more likely |
| Fast mode speed | Standard | 2.5x faster |
| Fast mode cost | Baseline | 3x cheaper |
| Dynamic Workflows | No | Yes (research preview) |
| Effort Control | No | Yes |
| Mid-task system prompts | No | Yes (API) |
| Price | $5/$25 per M tokens | Same |
Pricing and Access
Opus 4.8 pricing is unchanged from Opus 4.7:
| Mode | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| Standard | $5.00 | $25.00 |
| Fast mode | $10.00 | $50.00 |
The model is available now via the Claude API using the model string claude-opus-4-8. It’s also the default model powering Claude.ai for Opus users.
What About Claude Mythos?
Anthropic has also been quietly testing a more powerful model class called Claude Mythos with a small number of organizations.
Opus 4.8’s alignment profile is now described as “similar to Claude Mythos Preview” — which suggests the gap between the public flagship and Anthropic’s experimental top-tier model is narrowing.
Anthropic says it expects to bring Mythos-class models to all customers “in the coming weeks.” This will be worth watching.
Who Should Care About Claude Opus 4.8?
Developers and engineering teams will get the most obvious benefit from the Dynamic Workflows feature and the improved coding benchmarks. The ability to run codebase-scale migrations autonomously is a genuine step up.
Businesses using Claude for agentic tasks — financial analysis, legal research, knowledge work — will notice the judgment improvements. An AI that flags its own uncertainties rather than presenting flawed output confidently is fundamentally more useful in high-stakes environments.
Individual users on Claude.ai get the effort control feature, which makes Claude more flexible for everyday use without burning through rate limits on simple tasks.
API developers get mid-task system prompts, which opens up new possibilities for long-running workflows that need to adapt instructions dynamically.
FAQ
When was Claude Opus 4.8 released? Claude Opus 4.8 was announced and released on May 28, 2026.
Is Claude Opus 4.8 free? Opus 4.8 is available on Claude.ai for users on plans that include Opus access. API access is priced at $5 per million input tokens and $25 per million output tokens — same as Opus 4.7.
What is the API model string for Opus 4.8? Developers can access Claude Opus 4.8 via the API using claude-opus-4-8.
What is Dynamic Workflows? Dynamic Workflows is a new Claude Code feature that lets Claude plan large-scale tasks and run hundreds of parallel subagents in a single session. It’s in research preview for Enterprise, Team, and Max plan users.
How does Effort Control work? Users on claude.ai and Cowork can set how much thinking Claude puts into each response. Higher effort means deeper reasoning and better output. Lower effort means faster replies and slower rate limit usage.
Is Opus 4.8 better than GPT-5.5? Opus 4.8 outperforms GPT-5.5 on several benchmarks including SWE-Bench Pro (agentic coding), multidisciplinary reasoning, and computer use tasks. GPT-5.5 leads on terminal-coding benchmarks specifically.
What is Claude Mythos? Claude Mythos is Anthropic’s most advanced model class, currently in limited testing with select organizations. Anthropic says Mythos-class capabilities are coming to all customers “in the coming weeks.”
Does Opus 4.8 cost more than Opus 4.7? No. Pricing is identical. Standard use is $5/$25 per million tokens (input/output). Fast mode is $10/$50.
What platforms support Opus 4.8? Opus 4.8 is available on Claude.ai, the Claude API (via Anthropic), and through partners including Amazon Bedrock, Google Vertex AI, and Claude Code.
How is Opus 4.8 more honest than previous models? The model is four times less likely to let code flaws pass without flagging them, more likely to express uncertainty when warranted, and less likely to make unsupported claims. Anthropic’s alignment team also confirmed lower rates of deceptive behavior compared to Opus 4.7.
Final Thoughts
Claude Opus 4.8 is a meaningful release, not just a routine version bump. The honesty improvements alone change the nature of working with the model on real tasks — you spend less time second-guessing outputs and more time actually shipping things.
The Dynamic Workflows feature is potentially transformative for large-scale engineering work. And the pricing staying flat while performance improves across the board is exactly what you want from a model update.
Anthropic is also clearly building toward something bigger with Mythos on the horizon. In the meantime, Opus 4.8 is the most capable, most trustworthy version of Claude yet.
It’s available now.
