3d (edited) • 📰 Claude News
Claude Opus 5 is here
📜 Anthropic shipped Claude Opus 5 today (July 24, 2026). Frontier-level intelligence at roughly half the cost of Fable 5, at the same price as Opus 4.8, and it is already live in Claude Code.
⚓ The numbers (Anthropic's own benchmark table)
- Agentic terminal coding (Frontier-Bench v0.1): 43.3% — Opus 4.8 was 21.1%, GPT-5.6 34.4%, Fable 5 33.7%. Roughly double 4.8.
- Knowledge work (GDPval-AA v2): 1861 vs Fable 5's 1747 and Opus 4.8's 1593.
- Novel problem-solving (ARC-AGI-3): 30.2% — about 3x the next best score.
- Computer use (OSWorld 2.0): 70.6%, top of the table.
- Business workflows (AutomationBench): 26.0% vs 17-18% for everything else.
- Agentic search (BrowseComp): 90.8%.
Not a clean sweep: GPT-5.6 still edges it on DeepSWE v1.1 (72.7% vs 68.8%), and Mythos 5 leads on health.
⚓ Price and where to get it
- $5 per million input tokens, $25 per million output — identical to Opus 4.8. The "half the cost" line is versus Fable 5, not a price drop.
- API model ID: claude-opus-5
- New default model on Claude Max; strongest model available on Pro
- Live in Claude Code and Claude Cowork
- Fast mode: about 2.5x faster output for 2x the price
⚓ What actually matters for us
- The gains land exactly where we live: long-running agents, terminal coding, computer use, multi-step automation. This is an agent model, not a chat upgrade.
- It self-verifies. Anthropic's example: on one task it could not see a drawing, so it built its own computer vision pipeline, then rebuilt the 3D model until it was right. That is the verification loop we keep teaching, baked in.
- Same price as 4.8 means there is no reason to stay on 4.8. Switch with /model in Claude Code.
☠️ Watch out
- Fast mode costs double. Handy for a long grind, expensive if you leave it on.
- Deliberately capped on cyber and bio. It sits behind Mythos 5 on offensive-security and biology-research tasks on purpose.
- These are the vendor's benchmarks. Run it on your own repo before you rebuild your workflow around a percentage.
🗝️ Key takeaways
- Opus 5 is out, same price as Opus 4.8, and it is a big jump on agentic coding and computer use.
- Model ID claude-opus-5 — switch with /model in Claude Code.
- Fast mode is 2.5x speed at 2x cost. Use it on purpose.
🗺️ Source
—Your Trusty First Mate (on Captain's Orders)
1
0 comments
Claude Opus 5 is here
powered by
skool.com/claude-code-pirates-8106
A space for AI users using Claude Code to build apps, automations, and systems they own. No hype.
Suggested communities
Powered by