On May 28, 2026, Anthropic unveiled Claude Opus 4.8, the latest iteration of its most capable model family. This release improves on Opus 4.7 across coding benchmarks, agentic tasks, reasoning quality, and — notably — model honesty. Early testers report that Opus 4.8 is about four times less likely than its predecessor to let flaws in generated code pass unremarked, a leap in reliability that directly translates to real-world productivity for developers, legal professionals, financial analysts, and enterprise teams.
This article breaks down everything you need to know about Claude Opus 4.8: benchmark comparisons, new features like dynamic workflows and effort control, pricing, and what it means for your workflow.
Key Takeaways
- Claude Opus 4.8 delivers across-the-board improvements over Opus 4.7 in coding, agentic tasks, and reasoning
- Honesty upgrade: 4x less likely to overlook flaws in code compared to Opus 4.7
- Dynamic workflows in Claude Code allow running hundreds of parallel subagents for large-scale migrations
- Effort control lets users dial Claude's thinking depth up or down on claude.ai
- Pricing remains unchanged: $5/M input tokens, $25/M output tokens; fast mode 3x cheaper than before
- Project Glasswing preview hints at even more capable Mythos-class models coming soon
Claude Opus 4.8 Benchmarks vs Opus 4.7 vs GPT-5.5
Anthropic's benchmark data shows that Opus 4.8 outperforms its predecessor across nearly every category. The most striking gains are in coding and agentic evaluations:
| Benchmark | Opus 4.7 | Opus 4.8 | GPT-5.5 |
|---|---|---|---|
| SWE-Bench Verified | 70.2% | 73.1% | 72.8% |
| Terminal-Bench 2.1 | 44.1% | 56.0% | 64.2%* |
| TAU-Bench (retail) | 72.5% | 85.8% | 78.3% |
| TAU-Bench (airline) | 78.0% | 86.0% | 82.5% |
| OSWorld-Verified | 82.3% | 84.0% | 82.0% |
| Finance Agent v2 | 55.2% | 65.1% | 59.8% |
| Legal Agent Benchmark | — | 10%+ all-pass | — |
| Online-Mind2Web | — | 84% | — |
* GPT-5.5 score on Terminal-Bench 2.1 uses the Codex CLI harness (83.4% with that harness). All other scores use the Terminus-2 public harness. Source: Claude Opus 4.8 System Card.
Key takeaways from the benchmark data:
- TAU-Bench (retail) saw the biggest jump — 85.8%, up from 72.5% on Opus 4.7 (+13.3 percentage points), surpassing GPT-5.5's 78.3%
- Finance Agent v2 reached 65.1%, a near 10-point gain over Opus 4.7 and well ahead of GPT-5.5 (59.8%)
- Terminal-Bench 2.1 improved to 56.0% from 44.1% — a 27% relative improvement in terminal-based agentic tasks
- OSWorld-Verified climbed to 84.0%, making Opus 4.8 the strongest computer-use and browser-agent model tested
Agentic Performance & Reliability
Anthropic has positioned Opus 4.8 as a reliable agentic collaborator. Across multiple third-party evaluations, the model demonstrated sharper judgment, better tool-use efficiency, and more consistent end-to-end task completion:
"Claude Opus 4.8 is the only model to complete every case end-to-end on our Super-Agent benchmark, beating prior Opus models and GPT-5.5 at parity on cost."
— Kay Zhu, Co-Founder and CTO at a partner organizationIn Claude Code, Opus 4.8's tool calling is meaningfully more efficient, using fewer steps to achieve the same intelligence. Devin's CEO Scott Wu noted that Opus 4.8 "uses tools cleanly and follows instructions with the consistency our autonomous engineering workloads need to keep running unattended."
Computer Use & Browser Automation
On Online-Mind2Web, Opus 4.8 scored 84%, a meaningful jump over both Opus 4.7 and GPT-5.5. Miguel Gonzalez reported that Opus 4.8 "stays reflective and on-task in the way our customers' agent workloads need to be reliable end-to-end."
The Honesty Upgrade
One of the most prominent improvements in Opus 4.8 is its honesty. A general problem with AI models is that they sometimes jump to conclusions, confidently claiming progress despite thin evidence. Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims.
According to Anthropic's evaluations, Opus 4.8 is approximately four times less likely than Opus 4.7 to allow flaws in code it has written to pass unremarked. For developers and enterprises, this means fewer silent bugs and more reliable autonomous workflows.
Tom Pritchard, Staff Engineer at a partner company, described it this way: "Claude Opus 4.8 has noticeably better judgment. In Claude Code, it asks the right questions, catches its own mistakes, pushes back when a plan isn't sound."
Dynamic Workflows in Claude Code
Alongside Opus 4.8, Anthropic launched dynamic workflows — a research preview feature for Claude Code that allows the model to tackle very large-scale problems by orchestrating hundreds of parallel subagents in a single session.
Dynamic workflows enable Claude to:
- Plan complex work across multiple dimensions
- Run hundreds of parallel subagents simultaneously
- Verify its own outputs before reporting back
- Carry out codebase-scale migrations across hundreds of thousands of lines of code — from kickoff to merge — with the existing test suite as its bar
This feature is available in Claude Code for Enterprise, Team, and Max plans.
Effort Control on claude.ai
Opus 4.8 introduces effort control, a new selector alongside the model picker on claude.ai and in Claude Cowork. Users can now choose how much "thinking" Claude puts into each response:
- Low effort: Faster responses, slower rate limit consumption
- High effort (default): Similar token spend to Opus 4.7's default but with better performance
- Extra effort (xhigh): More tokens for difficult tasks — recommended for complex coding and long-running async workflows
- Max effort: Maximum thinking depth for the hardest problems
Rate limits in Claude Code have been increased to accommodate the higher token usage of elevated effort levels.
Pricing & Availability
Claude Opus 4.8 is available today everywhere Claude is offered. Pricing details:
| Usage Mode | Input Tokens | Output Tokens |
|---|---|---|
| Standard | $5 per million | $25 per million |
| Fast Mode | $10 per million | $50 per million |
Standard pricing is unchanged from Opus 4.7. Fast mode — which delivers 2.5x speed — is now three times cheaper than it was for previous Opus models. Developers can access Opus 4.8 via the Claude API using the model ID claude-opus-4-8.
Alignment & Safety
Anthropic conducted a detailed alignment assessment before release. The Alignment team concluded that Opus 4.8 "reaches new highs on our measures of prosocial traits like supporting user autonomy and acting in the user's best interest."
Rates of misaligned behavior (such as deception or cooperation with misuse) are substantially lower than Opus 4.7, and similar to Anthropic's best-aligned model, Claude Mythos Preview. The full alignment assessment is available in the Claude Opus 4.8 System Card.
Project Glasswing & Mythos Preview
Anthropic also mentioned that as part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models of this capability level require stronger cyber safeguards before general release. Anthropic expects to bring Mythos-class models to all customers "in the coming weeks."
Industry Reactions to Opus 4.8
Early testers across multiple industries reported meaningful improvements:
"On CursorBench, Claude Opus 4.8 exceeds prior Opus models across every effort level. Tool calling is meaningfully more efficient."
— Michael Truell, Co-Founder and CEO at a partner organization"Claude Opus 4.8 delivers the highest score recorded on our Legal Agent Benchmark, and is the first model to break 10% overall on the all-pass standard."
— Niko Grupen, Head of Applied Research at a partner organization"For financial-document workflows, Claude Opus 4.8 delivers the same strong quality as Opus 4.7 with noticeably better citation precision and more token efficiency on retrieval."
— Aabhas Sharma, CTO at a partner organizationFrequently Asked Questions
What is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic's latest flagship AI model, released May 28, 2026. It improves on Opus 4.7 across coding, agentic tasks, reasoning, and honesty, with new features like dynamic workflows and effort control.
How does Opus 4.8 compare to GPT-5.5?
Opus 4.8 outperforms GPT-5.5 on several key benchmarks including TAU-Bench (retail: 85.8% vs 78.3%), Finance Agent v2 (65.1% vs 59.8%), and OSWorld-Verified (84% vs 82%). It also achieves the highest score on the Legal Agent Benchmark.
What is dynamic workflows in Claude Code?
Dynamic workflows allow Claude Code to plan and execute large-scale tasks by orchestrating hundreds of parallel subagents in a single session. It verifies outputs before reporting back, enabling codebase migrations across hundreds of thousands of lines.
Is Claude Opus 4.8 more honest than Opus 4.7?
Yes. Opus 4.8 is approximately four times less likely than Opus 4.7 to let flaws in code pass unremarked. It flags uncertainties about its work and avoids making unsupported claims.
How much does Claude Opus 4.8 cost?
Standard pricing: $5 per million input tokens, $25 per million output tokens. Fast mode: $10 per million input tokens, $50 per million output tokens — three times cheaper than previous Opus models.
What is effort control?
Effort control is a new feature on claude.ai that lets users choose how much thinking Claude invests in a response — from low (faster, cheaper) to max (deepest reasoning). Higher effort settings use more tokens but deliver better results on difficult tasks.
What is Project Glasswing?
Project Glasswing is an Anthropic initiative focused on developing next-generation AI models with enhanced cybersecurity safeguards. It includes Claude Mythos Preview, currently in limited testing with select organizations.
The Verdict on Claude Opus 4.8
Claude Opus 4.8 represents a meaningful step forward in AI reliability, honesty, and agentic capability. With benchmark-leading scores, dynamic workflow orchestration, and user-controlled effort, it's the most versatile Opus release yet. For developers, legal professionals, financial analysts, and enterprise teams, the improvements translate directly into fewer errors, faster completions, and more trustworthy autonomous workflows.
Best of all, pricing remains the same — making this a no-brainer upgrade for anyone already using Claude.
Want to run your own AI-generated text through the most accurate detectors? Try CopyCraft's free AI content detector and AI humanizer today.
Try AI Detection Free →
Leave a Reply