Anthropic shipped Claude Opus 4.8, and if you write code for clients, this one matters.
Here's what actually changed:
→ 1M context window. Your whole repo fits. No more spoon-feeding files one by one.
→ Adaptive thinking. It spends more compute on hard problems, less on easy ones. You stop paying for overthinking on a one-line fix.
→ It catches its own mistakes. Production-ready code with way less babysitting.
→ 84% on Online-Mind2Web — best browser-agent model tested. The "AI does the boring tasks" future is closer than the hype suggested.
Pricing: $5 in / $25 out per million tokens. With prompt caching you cut up to 90% of that.
For freelancers this is leverage, not a threat. The dev who pairs with this ships 3x the work at the same quality. The one who ignores it competes on hours.