Claude Opus 5 Ships: Same Price as 4.8, Double the Agentic-Coding Score
Anthropic released Claude Opus 5, succeeding Opus 4.8 as the flagship Opus-tier model. Pricing stays flat at $5/$25 per million input/output tokens — the same as its predecessor — while delivering what Anthropic and third-party benchmarks describe as near-Fable-5-level performance at roughly half the cost. It ships with a 1M-token context window and 128K max output, and is now the default model on Claude Max and the strongest available option on Claude Pro.
New: an effort toggle
Opus 5 introduces a low/medium/high effort toggle, letting you trade cost for capability on a per-request basis rather than being locked into a single quality/cost tradeoff for the whole model. A Fast mode variant doubles the rate ($10/$50 per million tokens) and runs roughly 2.5× faster — useful when latency matters more than minimising cost for a specific request.
Benchmark highlights
- Frontier-Bench v0.1 (third-party agentic coding): 43.3% — more than double Opus 4.8's 21.1%, and ahead of both Fable 5 (33.7%) and GPT-5.6 Sol (34.4%).
- CursorBench 3.2: within 0.5% of Fable 5's peak score, at half the cost per task.
- ARC-AGI 3: roughly three times the next-best model's score.
- OSWorld 2.0: surpasses Fable 5's best result at just over a third of the cost.
Opus 5 meaningfully changes the earlier "Sonnet 5 vs. Opus 4.8" calculus from earlier this month (see July 5's entry): if you'd been routing hard agentic-coding tasks to Sonnet 5 purely on cost grounds, Opus 5's flat pricing versus 4.8 combined with the Frontier-Bench jump is worth re-testing against — you may now get meaningfully better results at the same Opus-tier price you were already budgeting for.