Claude Opus 4.8 Lands Everywhere: The New Flagship That Misses 4x Fewer Code Flaws!

Hi everyone, it's Shiichan! I have some big news to share today, and I've been buzzing about it all morning. There's a brand-new model in town!

Anthropic News anthropic.com

What was announced?

Over on Anthropic's News, the new flagship model Claude Opus 4.8 was announced! It builds on the previous Opus 4.7 and raises the bar across a bunch of benchmarks. The best part: standard pricing stays the same, at $5 per million input tokens and $25 per million output tokens. It's available across every platform already, and in the API you call it with the claude-opus-4-8 identifier.

The story so far

Opus 4.7 was already great, but on code reviews and long agentic tasks there were still moments where you wished it were a little sharper and a little more persistent. Opus 4.8 is all about pushing on exactly those areas.

What changes

First, something developers will love: code quality. Opus 4.8 is about 4x less likely to let code flaws pass unremarked compared to the previous model. As a review buddy, it just got a lot more dependable.

The agentic skills leveled up too. On the computer-use benchmark Online-Mind2Web it scored 84%, surpassing both Opus 4.7 and GPT-5.5. It's also the first model to exceed 10% on the Legal Agent Benchmark all-pass standard. And on the coding benchmark CursorBench, it outperforms prior Opus models across all effort levels.

Dive Deep

There are some new features too.

Dynamic workflows (a Claude Code research preview) let Claude run hundreds of parallel subagents in a single session. It can carry out codebase-scale migrations across hundreds of thousands of lines of code, from kickoff all the way to merge. That's a huge scale!

Effort control is a new control next to the model selector. You choose how much effort Claude puts into a response, so you can balance answer quality against token usage yourself. It's available in claude.ai and Cowork, on all plans.

For developers, the Messages API got handier too. You can now add a system entry mid-conversation to update Claude's instructions without breaking the prompt cache or routing the update through a user turn. Super helpful when you want to nudge the plan during a long agentic task.

A bit more on pricing: fast mode runs at 2.5x the speed, at $10 per million input tokens and $50 per million output tokens, and it's now three times cheaper than fast mode was for previous models. Reaching for speed just got a lot more realistic.

On safety, the model hits new highs on prosocial traits, and its rate of misaligned behavior is substantially lower than Opus 4.7.

Wrap-up

• Anthropic's News announced the new flagship Claude Opus 4.8, available on every platform now (API id claude-opus-4-8)
• Standard pricing is unchanged at $5 input / $25 output per million tokens
• ~4x less likely to miss code flaws, 84% on Online-Mind2Web, and first past 10% on the Legal Agent Benchmark all-pass standard
• Dynamic workflows bring hundreds of parallel subagents, effort control lets you pick the effort level, and the Messages API now accepts mid-task system entries
• Fast mode runs 2.5x faster and is 3x cheaper than before

If you're an engineer who wants to hand code review and big refactors to an agent, or someone who likes fine-tuning the balance of smarts, speed, and cost, this is the update that'll hit home for you!