Anthropic ships Claude Opus 5
Opus 5 doubles Opus 4.8's Frontier-Bench score at the same price, triples the next-best model on ARC-AGI 3, and adds mid-conversation tool switching.
Anthropic released Claude Opus 5 on July 24, holding Opus 4.8’s pricing of $5 per million input tokens and $25 per million for output, available immediately on the Claude API, Claude.ai, Claude Max and Pro, and now Amazon Bedrock. Anthropic says Opus 5 doubles Opus 4.8’s score on its Frontier-Bench coding evaluation at lower cost, scores three times higher than the next-best model on ARC-AGI 3, and beats Fable 5 on the OSWorld 2.0 computer-use benchmark at roughly a third the cost. Fast mode runs 2.5x faster at double the price.
Beyond raw benchmarks, Anthropic cites gains in scientific reasoning: 10.2 points on organic chemistry tasks, 7.7 points on protein sequence prediction, plus stronger verification on multi-step problems. Two new beta features ship with the model: mid-conversation tool changes, so a running agent session can swap tools without restarting, and automatic fallback routing when a request gets blocked. On alignment testing, Anthropic reports Opus 5’s lowest misaligned- and deceptive-behavior scores of any recent Claude model, though it still trails Mythos 5 on cybersecurity exploitation tasks. Devin’s CEO said the model “approaches Fable-level performance at half the cost”; Box’s CTO cited an 8% platform-wide improvement with 17% gains on due-diligence workflows.
The pricing hold matters most for production workloads: Anthropic is not charging more for what it says is roughly double the coding capability, raising the bar Opus has to clear against cheaper Sonnet-tier routing. It lands the same week AMD committed up to $5 billion and 2GW of chip capacity to Anthropic, and as Claude and ChatGPT both shipped voice-agent upgrades: Anthropic is pushing model quality, infrastructure, and interface at once. If workloads are routed to Sonnet purely for cost, this is the point to re-run your evals against Opus 5 before assuming the price gap still beats the capability gap.