Anthropic ships Claude Opus 5.5: Frontier agentic coding at a 40% discount
Opus 5.5 matches Fable 5.1 performance but slashes costs by 40%. It beats GPT-6 Astra on agentic coding for a fraction of the price.
Models · Source: Hacker News
What happened
Anthropic just dropped Claude Opus 5.5. It is their first major release since they publicly called to pace frontier AI development. The model performs at the level of Claude Fable 5.1 but costs 40 percent less to run than Opus 5. It also generates output 30 percent faster and communicates more naturally.
The benchmark results focus heavily on agentic coding and efficiency. Opus 5.5 beats GPT-6 Astra on FrontierCode at roughly 20 percent of the cost per task. On Terminal Bench 4.0, it matches Astra for about 40 percent of the cost. One early tester used it to complete a 680,000-line code migration in under a day, while another audited a 200,000-line codebase in under three hours.
Anthropic is also locking down security for high-risk domains. Opus 5.5 requires vetted access through their Life Sciences and Cyber Verification Programs for biology and cybersecurity work. It features a pre-action classifier and stronger defenses against prompt injection attacks. Anthropic also announced that Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks.
Key facts
- 40% — Cost reduction compared to Opus 5 on typical workloads.
- $0.20 — Cost per million tokens for cache reads, a 60 percent drop.
- 680,000 — Lines of code migrated by a tester in less than a day.
- $4 — Cost per million input tokens.
- $20 — Cost per million output tokens.
- 11 — Points by which Opus 5.5 beats GPT-5.6 Sol on CursorBench.
Why it matters
The unit economics of agentic workflows just shifted massively. Cache reads now cost a mere $0.20 per million tokens. This 60 percent price drop from Opus 5 means you can finally run long-context, multi-step autonomous agents without burning through your runway. Builders can deploy agents that run unattended for hours to translate massive codebases from C to Rust or handle complex migrations cheaply. The era of agonizing over token costs for background agent tasks is ending.
The AI race is moving from raw intelligence to cost-efficiency and safety. Anthropic is proving you do not need to pay a premium for frontier capabilities if the model is optimized for agentic tasks. But the strict verification programs for biology and cyber tasks show the regulatory walls are closing in. If you build in sensitive sectors, expect heavy gatekeeping on model access. You will need to build compliance into your roadmap from day one.
For builders
Leverage ultra-cheap context caching
Cache reads now cost $0.20 per million tokens. You should immediately rewrite your agentic loops to maximize cache hits. Startups building heavy RAG or coding agents win big here while competitors burn cash.
Prepare for gated model access
Anthropic is gating biology and cybersecurity capabilities behind strict verification programs. If your product touches these verticals, you must apply for access now or risk losing your core engine. Unverified startups will lose to incumbents who pass the vetting.
Use Fast mode for latency apps
Opus 5.5 offers a Fast mode with up to 2.5x speed for $8 per million input tokens. Consumer apps where users wait for responses should eat this cost. You pay double the base input price, but you buy a vastly better user experience.
My take
Anthropic is playing a brilliant game by commoditizing agentic coding while OpenAI chases raw scale. I love that they slashed cache read prices by 60 percent, because that is where real builders actually spend their money. Gating cyber and biology behind verification is annoying, but it is the exact move you make to avoid government crackdowns.
Original reporting: Hacker News. This is my rewrite and opinion.