Anthropic released Claude Opus 4.8 today. It builds on Opus 4.7 with the same price but better performance. This article covers the five changes that matter most for developers.
1. It Catches Its Own Bugs — 4x More Often
This is the biggest change in Opus 4.8.
Previous versions would sometimes complete a task without mentioning problems in the code. Opus 4.8 is four times more likely to flag issues it finds in its own output.
In practice, this means Claude will tell you when something looks wrong — instead of silently finishing and leaving you to find the bug later.
Anthropic calls this an honesty improvement. The model is more likely to say “I’m not sure this is correct” or “there is a potential issue here.”
2. Dynamic Workflows — Hundreds of Parallel Agents
Claude Code now supports dynamic workflows.
You give Claude a large task. Claude plans the work, then runs hundreds of parallel subagents to complete it. Each subagent works on a separate part. Claude checks the results and merges them.
The result: Claude Code can now handle entire codebase migrations from start to finished pull request — without you managing each step.
This is useful for:
- Migrating a large codebase to a new framework
- Refactoring hundreds of files at once
- Running automated tests across a large codebase
3. Effort Control — You Choose How Hard It Thinks
Opus 4.8 adds an effort control setting.
You can set the effort level to high, extra, or max. Higher effort means Claude thinks longer and uses more tokens. It produces better results on complex tasks.
In the API, you control this with the thinking parameter:
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(
model="claude-opus-4-8",
max_tokens=16000,
thinking={
"type": "enabled",
"budget_tokens": 10000 # increase for harder tasks
},
messages=[{
"role": "user",
"content": "Review this code and find all edge cases."
}]
)
In Claude Code, use --effort xhigh for difficult tasks. In claude.ai, a dropdown lets you pick the effort level.
The default is high — a good balance of quality and speed.
4. Fast Mode Is Now 3x Cheaper
Fast mode runs Claude at 2.5x the normal speed. In Opus 4.7, fast mode was expensive. In Opus 4.8, fast mode costs three times less.
This makes fast mode practical for high-volume tasks — code reviews, document parsing, batch processing.
Standard pricing stays the same:
- Input: $5 per million tokens
- Output: $25 per million tokens
Fast mode pricing dropped significantly with this release.
5. Benchmark Results
Opus 4.8 outperforms Opus 4.7 across key benchmarks:
| Benchmark | Opus 4.7 | Opus 4.8 |
|---|---|---|
| Agentic coding | 64.3% | 69.2% |
| Reasoning with tools | 54.7% | 57.9% |
Opus 4.8 also beats GPT-5.5 and Gemini 3.1 Pro on agentic coding and reasoning tasks. OpenAI’s model still leads on agentic terminal coding specifically.
How to Upgrade
Upgrading is a one-line change in your API calls:
# Before
model="claude-opus-4-7"
# After
model="claude-opus-4-8"
The model is available now on claude.ai, the Claude API, and AWS Bedrock.
What Is Coming Next
Anthropic hinted at a broader release called “Mythos” coming in the next few weeks. No details yet, but it suggests larger changes beyond a point release.