Anthropic Ships Claude Opus 5, Claims It Matches Flagship Fable 5 on Coding at Half the Cost

Anthropic released Claude Opus 5 on July 24, claiming the model matches or exceeds its flagship Fable 5 on most coding and knowledge-work benchmarks at roughly half the price, according to the company. Opus 5 is priced at $5 per million input tokens and $25 per million output tokens — the same rate as its predecessor.

The launch marks the fourth model Anthropic has shipped in under two months, following Fable 5 in early June and Sonnet 5 at the end of June. That cadence is a sharp break from Anthropic's historical pattern of roughly one flagship release per quarter.

Benchmark claims

According to Anthropic, Opus 5 more than doubles its predecessor's score on Frontier-Bench, a coding evaluation that tests whether models can build working software directly from engineering drawings, and surpasses every competing model on that test, including Fable 5. The company also claims Opus 5 scores roughly three times higher than the next-best model on ARC-AGI 3, a benchmark for novel problem-solving, and outperforms rivals on Zapier AutomationBench — a test of whether models can complete real business tasks end to end — even when run at its lowest effort setting.

None of these figures have been independently verified, and Anthropic has not published exact numerical scores alongside the relative comparisons.

Alignment claims

Anthropic says Opus 5 is the most aligned model it has shipped, recording the lowest rate of deceptive behavior in its internal automated safety audit. The company says this builds on Fable 5, which set what Anthropic described as a new honesty benchmark at its own May launch. Opus 5 reportedly still trails Mythos 5 on offensive cybersecurity tasks — a gap Anthropic attributes to a deliberate choice not to train Opus 5 on cyber-offense work, though it notes the model's cybersecurity performance improved as a side effect of general capability gains.

Availability and new features

Opus 5 is now the default model on Claude Max and the top-tier option on Claude Pro. Developers can reach it through the Anthropic API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. A Fast mode runs at two and a half times default speed for double the base price.

Two API features ship alongside the model: mid-conversation tool switching that lets developers change available tools without invalidating the prompt cache, and automatic fallback routing that redirects flagged requests to a different model instead of blocking them outright.

What this means

Four model releases in eight weeks signals Anthropic is shifting from a single-flagship strategy to a tiered lineup addressing distinct price-performance points — Opus 5 now sits between the premium Fable tier and budget-oriented Sonnet tier. The unverified benchmark claims and vague alignment metrics (no absolute deception rate was disclosed) make it difficult to assess real-world gains until third-party evaluations arrive. Keeping pricing flat while claiming a coding leap over the previous Opus generation suggests Anthropic is prioritizing volume adoption ahead of OpenAI and Google's next releases rather than extracting margin from the upgrade.