FAST TAKE · 2026-07-24 · ANTHROPIC CLAUDE OPUS 5
Claude Opus 5 — same price, materially better
Anthropic shipped Claude Opus 5 on July 24 at $5 per million input and $25 per million output — identical to Opus 4.8, which it replaces. Anthropic positions it as approaching Fable 5 intelligence at half the cost, and reports it as their most aligned model to date on internal behavioural audits. Hosted-only, so no local pick changes; we have renamed the Opus row in the cost calculator rather than adding one, which is what we have done for every Opus generation since 4.6.
Verdict: Near-Fable-5 capability at half the price, and the price did not move from Opus 4.8 — a straight upgrade for anyone already on the Opus row
The take
Verified against Anthropic's own announcement and the platform rate card: "Claude Opus 5 is available today on all platforms, priced at $5 per million input tokens and $25 per million output tokens (the same as Opus 4.8)." The pricing page lists it alongside Opus 4.8 at identical rates, with Fast mode at $10/$50. Anthropic's claims: new state of the art on Frontier-Bench and GDPval-AA, within 0.5% of Fable 5's peak CursorBench 3.2 score at half the cost per task, and beating Fable 5's best OSWorld 2.0 result at just over a third of the cost. On alignment they report it adhering to the constitution better than Opus 4.8, Sonnet 5 or Fable 5, with the lowest rates of deceptive behaviour they have measured.
The rename, not add, is deliberate and worth stating since it is the third time we have done it. The cost calculator exists to answer "should I buy hardware or pay for this," and that question is always about the flagship you would actually provision today — not a museum of every Opus version. Price identical, capability higher, predecessor superseded: the row becomes Opus 5. Same call we made for 4.6 → 4.7 → 4.8. The historical fast takes stay where they are as a dated record.
Two things we deliberately did not do. Claude Mythos 5 is on the rate card at $10/$50 but is limited-availability, and we do not put models you cannot reliably provision into a budgeting tool — the same reasoning that kept Fable 5 out while it was suspended in June. And nothing here touches a local pick, because Opus is hosted-only; if you are choosing between an RTX 5090 and an Opus subscription, the relevant number is that Opus 5 costs the same as Opus 4.8 did, so your break-even month has not moved. What has moved it, this same week, is OpenAI cutting Luna by 80% — see that take for the arithmetic.
Where this fits
Models: GLM-5.1 · Kimi K3 (2.8T-A50B) · DeepSeek V4-Pro · Qwen 3.6-35B-A3B
Hardware: NVIDIA RTX 5090 · Mac Studio M3 Ultra 96 GB · NVIDIA DGX Spark
Sources
Next step
Try this in the planner→