What happened
On July 24, 2026, Anthropic released Claude Opus 5, available the same day through the Claude API (model ID claude-opus-5), Claude.ai, Claude Code, and Claude Cowork. It is now the default model on the Claude Max subscription, and the top option offered on Claude Pro. A faster mode runs roughly 2.5 times quicker at double the base price.
Pricing holds at $5 per million input tokens and $25 per million output tokens, the same rate Anthropic charged for the prior Opus 4.8. Anthropic frames Opus 5 as coming close to the frontier intelligence of Claude Fable 5, its higher-priced flagship from June, at half the price. That framing is Anthropic's own and does not hold evenly: on benchmarks it names, such as CursorBench 3.2 and OSWorld 2.0, Anthropic says Opus 5 lands within a fraction of a percent of, or ahead of, Fable 5's results at lower cost, while other benchmarks give no Fable 5 comparison. Anthropic also states Opus 5 remains behind Mythos 5, the separate, access-restricted model covered here previously, specifically on cybersecurity tasks such as exploit development. The three are distinct models with different capability profiles, not interchangeable versions of one another.
Anthropic's other reported figures, all from its own announcement and unverified here: a Frontier-Bench v0.1 score more than double Opus 4.8's at lower cost per task, an ARC-AGI 3 score three times the next-best model, and double-digit percentage-point gains over Opus 4.8 on life-sciences evaluations like organic chemistry and protein prediction. On safety, Anthropic says an internal audit scored Opus 5 at 2.3 for overall misaligned behavior, the lowest of its recent models, and that it follows Claude's Constitution better than Opus 4.8, Sonnet 5, or Fable 5, while still trailing Mythos 5 on cybersecurity evaluations. TechCrunch also reported Opus 5 skips the 30-day data retention policy applied to Fable 5 and Mythos 5, and that safety classifiers are expected to trigger about 85% less often than for Fable 5.
Why it matters
Anthropic is no longer selling one flagship model at one price; Opus, Fable, and the gated Mythos now sit at different price and capability points, so choosing a model is about matching a tier to a task and budget, not picking the newest release. Holding Opus 5's price at the Opus 4.8 rate while claiming better results, if accurate, would make agentic coding and long, iterative tasks cheaper to run at no extra cost. But every capability and safety figure here comes from Anthropic's own testing, not an outside evaluator, and Anthropic's account already shows Opus 5 losing ground to Mythos 5 in at least one area, which argues against reading "close to Fable 5" as general equivalence.
What to do next
- Do not assume Opus 5 replaces Fable 5 or Mythos 5 outright; Anthropic's own comparisons are benchmark-specific, and it says Opus 5 still trails Mythos 5 on cybersecurity tasks.
- If you already pay for Opus 4.8 usage, the price is unchanged, so any capability gain applies at no extra cost once you verify it on your own workload.
- Wait for independent benchmarks before deciding based on Anthropic's self-reported scores.
- For security research or exploit-adjacent work, do not assume Opus 5 matches Mythos 5's reported capability.
- Check the data-retention terms for your model and plan; Anthropic reportedly treats retention differently across Opus 5, Fable 5, and Mythos 5.
This briefing summarizes Anthropic's July 24, 2026 announcement and TechCrunch's same-day report, and cites The Verge's same-day coverage as a third dated source. All benchmark, safety, pricing, and efficiency figures are Anthropic's own claims as reported, not independently verified by LumoMate.