Anthropic Ships Claude Opus 5: Tops FrontierBench, Costs Half as Much as Fable 5
Anthropic just shipped its fourth flagship model in under two months — Claude Opus 5, which launched on July 24 and topped multiple benchmarks while keeping the same price as Opus 4.8.
Three Key Takeaways
Benchmark dominance across the board. On FrontierBench v0.1 (agentic coding), Opus 5 scored 43.3% at maximum effort, beating OpenAI's GPT-5.6 Sol at 37.5%. On ARC-AGI-3 (novel reasoning), it hit 30.2% — roughly three times the runner-up. On the GDPval-AA v2 knowledge-work evaluation, Opus 5 reached an Elo of 1,861, the current high-water mark. On CursorBench 3.2, Opus 5 performed within 0.5% of Fable 5's peak at half the cost per task.
Surgical pricing. Opus 5's API price is $5 input / $25 output per million tokens — unchanged from Opus 4.8 and exactly half of Fable 5's $10 / $50. The model supports a 1-million-token context window with up to 128K output tokens and offers a low/medium/high effort toggle, letting users balance cost against capability on every call.
Four flagships in two months. Mythos 5, Fable 5, Sonnet 5, and now Opus 5 — Anthropic's release cadence has gone from quarterly to biweekly. Opus 5 was available on day one across the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, claude.ai, Claude Code, and Cowork.
WangDou's Take
Four flagships in two months — Anthropic is shipping less like a research lab and more like a production line. But the numbers hold up: Opus 5's 43.3% on FrontierBench versus GPT-5.6 Sol's 37.5% is nearly a six-point lead; its ARC-AGI-3 score is triple the next model — that's not a rounding error, that's a generational gap. The killer move is pricing: near-identical performance to Fable 5, but at half the cost. That turns the buying decision into a simple math problem — unless you have a specific reason to need Fable, nobody is going to pay double for a 0.5% edge. Anthropic is running the "ceiling performance, floor price" playbook, and that puts real pressure on both OpenAI's and Google's pricing structures.
Source: TechCrunch, Fortune, Axios