What matters most in Claude Opus 4.8 isn't speed — it's honesty

There's a new Claude model out. Anthropic released Opus 4.8 on May 28. The press will write about benchmarks, speed, and how many agents it can run in parallel. That's not what matters most for you as an SMB leader.

What matters most is that the model is more honest about what it doesn't know.

Anthropic's own numbers: Opus 4.8 is about four times less likely to let code flaws pass unremarked compared to the previous version (4.7). And it produces seventeen times less dishonest summaries of its own work compared to Sonnet 4.6.

Why is that the only benchmark that counts for an SMB?

AI that pretends to know something it doesn't is costing you money and trust. A model that says "I couldn't find anything on this" is safe to use. A model that hallucinates an answer you'll have to fact-check anyway isn't a time-saver — it's a new step in your workflow. The difference between the two isn't a matter of taste. It's the difference between a tool that actually works day-to-day and one that requires constant supervision.

It's also the part that's most under-reported in every AI release. While the industry press writes about speed and parameter counts, the quiet error rate is what decides whether a business can take the tool seriously.

The other things worth noting:

• ▸Pricing is unchanged: $5 per million input tokens, $25 per million output tokens. If you already have a cost model, nothing shifts.
• ▸"Fast mode" — the mode where the model works about 2.5x faster — is now three times cheaper than on 4.7. For Claude Code users who've turned this on (`/fast`), that's real money on the bill.
• ▸"Dynamic workflows" is the new agent feature that lets Claude plan larger jobs and run hundreds of parallel sub-agents. It's NOT available on the Pro plan — only Enterprise, Team, and Max. If you're a solo user on Pro, you don't get this. If you have Claude Code via a company subscription, you do.

What does it mean practically?

For 90% of Claude users in Norwegian SMBs, nothing visible changes. The model is where it was. It's a bit better. It costs the same. You don't need to do anything — Claude.ai and Claude Code have already updated automatically.

The 10% you should check:

1. If you pay for the API directly — your own automation flows, an n8n workflow, or integration in your own tool — swap the model name to `claude-opus-4-8`. Same price, better quality. 2. If you use Fast mode in Claude Code, you save two-thirds on price for the same work. Good for cost management on longer projects. 3. If you do agent work — tools that plan and execute multi-step jobs themselves — consider dynamic workflows if you're on Enterprise or Team. That's where the biggest lift is for autonomy.

A small note on consultant-hunting:

If you're hiring an AI consultant and want to assess if they're any good: ask which model version they're building on, and why. A consultant who doesn't know what's changed in the past four weeks has stopped paying attention. One who knows AND can explain what the change means for the tool they're delivering to you — you can take seriously.

That's what Opus 4.8 changes most. Not speed. Not capability. Trust.