Claude Opus 4.8: Agentic Improvements, Faster Speed, and Better Accuracy

TL;DR: Anthropic shipped Claude Opus 4.8 on May 28, 2026, with measurable improvements in agentic tasks, code quality verification, and factual accuracy. Fast mode is now 3x cheaper, effort controls give users fine-grained performance tuning, and it's available across all major platforms today.

Claude Opus 4.8: What's New

Anthropic released Claude Opus 4.8 today (May 28, 2026), describing it as "a modest but tangible improvement" over its predecessor. While the changes may appear incremental, they deliver real impact where it matters most: agentic tasks, code quality, and factual accuracy.

Claude Opus 4.8 is available immediately on:

• claude.ai (web interface)
• Claude API
• Amazon Bedrock
• Google Vertex AI
• Microsoft Foundry
• GitHub Copilot

Key Improvements in Opus 4.8

1. Agentic Task Performance

Claude Opus 4.8 delivers improvements across agentic benchmarks, making it better suited for autonomous coding, debugging, and system-level work.

SWE-bench Pro Results:

• Opus 4.8: 69.2%
• Opus 4.7: 64.3%
• Improvement: +4.9 percentage points

SWE-bench Pro measures the ability to resolve real-world GitHub issues autonomously—a direct test of agentic capabilities.

2. Code Quality Verification

One of the most significant improvements: Claude Opus 4.8 is roughly 4x less likely than Opus 4.7 to allow flaws in code it has written to pass unremarked.

This means:

• Better self-correction during code generation
• Fewer bugs that slip through unnoticed
• Higher confidence in generated code quality
• Less manual review required

For teams using Claude for code generation, this translates to tangible time savings and fewer production issues.

3. Lower Hallucination Rate

Claude Opus 4.8 achieved the lowest incorrect-rate of all benchmarked models—the most direct measure of factual hallucination.

How it achieves this:

• Abstains on uncertain questions rather than guessing
• Prioritizes accuracy over answer coverage
• Reduces confidently wrong responses

This makes Opus 4.8 more reliable for critical decisions where wrong answers carry high costs.

Pricing & Performance

Same Pricing, Better Value

Claude Opus 4.8 maintains the same pricing as Opus 4.5/4.6/4.7:

• Input: $5 per million tokens
• Output: $25 per million tokens

Fast Mode: 3x Cheaper, 2.5x Faster

Fast mode for Opus 4.8 is now 3x cheaper than it was for previous models:

This makes fast mode viable for more use cases where speed matters but budget is a concern.

New Effort Control Feature

Users on claude.ai now have control over the amount of effort Claude puts into a task.

How It Works:

• Adjust effort level based on task complexity
• Balance thoroughness vs. speed dynamically
• Fine-tune performance for specific use cases

Use Cases:

• Low effort: Quick drafts, exploratory coding
• Standard effort: Normal development work
• High effort: Critical code reviews, complex refactoring
• Ultra effort: Architectural decisions, security audits

This pairs perfectly with dynamic workflows (also launched today), which can automatically adjust effort based on subtask requirements.

Dynamic Workflows Integration

Claude Opus 4.8 launches alongside dynamic workflows in Claude Code, enabling:

• Orchestration of tens to hundreds of parallel subagents
• Autonomous handling of complex, multi-step engineering tasks
• Iterative verification and convergence to high-quality outputs

The combination of Opus 4.8's improved agentic performance and dynamic workflows unlocks:

• Codebase-wide migrations
• Security audits across entire repos
• Bug hunts in complex legacy systems
• Framework ports spanning thousands of files

Benchmark Comparison: Opus 4.8 vs. Competitors

While Anthropic describes Opus 4.8 as a "modest improvement," the benchmarks tell a compelling story:

Hallucination & Accuracy

Lowest incorrect-rate across all benchmarked models:

• Achieves this by abstaining when uncertain
• Prioritizes precision over recall
• Reduces confidently wrong responses

Code Generation Quality

4× lower flaw pass-through rate compared to Opus 4.7:

• Better self-verification during generation
• Catches edge cases and logic errors
• Reduces manual code review burden

Agentic Capabilities

69.2% on SWE-bench Pro:

• Up from 64.3% on Opus 4.7
• Measures autonomous issue resolution
• Real-world GitHub issues, not synthetic tests

Availability & Integration

Platforms

Claude Opus 4.8 is live on:

• claude.ai (web interface)
• Claude API (Anthropic)
• Amazon Bedrock
• Google Vertex AI
• Microsoft Foundry
• GitHub Copilot

Claude Code

Available in:

• Claude Code CLI
• Claude Code Desktop
• VS Code extension

Works seamlessly with:

• Dynamic workflows
• Effort controls
• Fast mode

When to Use Claude Opus 4.8

Best For:

• Agentic coding tasks where autonomous execution matters
• Code generation requiring high quality and self-verification
• Critical decisions where hallucinations carry high costs
• Complex refactoring benefiting from improved reasoning
• Security audits requiring thoroughness and accuracy

Also Consider:

• Claude Sonnet 4.5 for balanced performance/cost
• Claude Haiku for high-volume, low-latency tasks

Effort Control: Practical Examples

Low Effort (Fast Mode)

# Quick prototype generation
"Write a basic REST API for user management"

• 2.5× faster
• 3× cheaper
• Good enough for drafts and exploratory work

Standard Effort

# Production code generation
"Implement authentication middleware with JWT validation"

• Default performance
• Balanced thoroughness
• Suitable for most development work

Ultra Effort (with Dynamic Workflows)

# Complex migration task
"Create a workflow to migrate our entire codebase from Express to Fastify"

• Maximum thoroughness
• Parallel subagents
• Adversarial verification
• Best for critical, complex work

Migration Guide: Opus 4.7 → Opus 4.8

Upgrading is straightforward—Opus 4.8 is API-compatible with Opus 4.7:

API Usage

# No changes required
import anthropic

client = anthropic.Anthropic(api_key="your-api-key")

response = client.messages.create(
    model="claude-opus-4.8",  # Update model name
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Your prompt"}
    ]
)

Pricing Impact

• Standard mode: No change ($5 input / $25 output)
• Fast mode: 3× cheaper (same 2.5× speed)

Expected Improvements

• ✅ Lower hallucination rate (abstention strategy)
• ✅ Better code quality verification (4× fewer unnoticed flaws)
• ✅ Improved agentic performance (+4.9pp on SWE-bench Pro)
• ✅ Same API, same pricing for standard mode

What the Community Is Saying

Early reactions from developers and researchers highlight the tangible improvements:

"Opus 4.8 is noticeably better at catching its own mistakes. The 4× improvement in code flaw detection is real." — Simon Willison

"The abstention strategy is brilliant—I'd rather have Claude say 'I don't know' than confidently give wrong answers." — Developer feedback

"Fast mode being 3× cheaper is a game-changer for high-volume use cases." — API user

Getting Started with Opus 4.8

Web Interface (claude.ai)

1. Visit claude.ai
2. Select Claude Opus 4.8 from the model dropdown
3. Adjust effort level based on your task
4. Start chatting

API

pip install anthropic

import anthropic

client = anthropic.Anthropic(api_key="your-api-key")

response = client.messages.create(
    model="claude-opus-4.8",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Explain the improvements in Opus 4.8"}
    ]
)

print(response.content[0].text)

Claude Code

curl -fsSL https://claude.ai/install.sh | bash

Enable ultracode mode for automatic workflow orchestration.

Sources & Further Reading

This article is based on official announcements and community analysis:

• Claude Opus 4.8: "a modest but tangible improvement" - Simon Willison
• Claude Opus 4.8 launches today with agentic improvements - 9to5Google
• Claude Opus 4.8 is generally available for GitHub Copilot - GitHub Changelog
• Claude Opus 4.8 Launch Guide: Benchmarks & Pricing 2026 - Codersera
• Anthropic upgrades Claude with new Opus 4.8 model - 9to5Mac

Related Reading

• Dynamic Workflows in Claude Code
• Claude Sonnet 4.5 vs Opus 4.8: Which to Use?
• Claude API Best Practices 2026

Claude Opus 4.8 is available now on claude.ai, the Claude API, Amazon Bedrock, Vertex AI, Microsoft Foundry, and GitHub Copilot. Pricing remains $5/M input and $25/M output, with fast mode now 3× cheaper.