Anthropic released Claude Opus 5 on Thursday, pricing it at $5 per million input tokens and $25 per million output tokens — the same as predecessor Opus 4.8 and half the price of the company's top-tier Fable 5 model. The company said Opus 5 "comes close to the capabilities of Claude Fable 5 in many domains" while handling complex coding tasks better than any prior Opus release. It arrives days after an OpenAI security incident dominated industry conversation and weeks after Anthropic's own standoff with the US government over Fable 5.
The Fable 5 episode reshaped the release calendar for frontier models. Anthropic's public-facing Mythos-class model drew federal cybersecurity concerns earlier this year, was pulled offline for a few weeks alongside Mythos 5, and came back with stronger safeguards. Those negotiations effectively opened a new phase of AI regulation under the Trump administration, and OpenAI felt the effects almost immediately: GPT-5.6 launched into a roughly two-week limited preview restricted to government-approved entities before broader release.
Anthropic is pitching Opus 5 as the everyday enterprise workhorse — best for knowledge work and biology use cases — while positioning Fable 5 as the model for the most ambitious long-horizon agent projects. The company emphasized in its release that Opus 5 carries stronger cybersecurity safeguards than Opus 4.8 and described it as "the most aligned Opus model and is the least susceptible to being tricked into misuse." That framing is deliberate given how the past several months have played out.
Key facts
- 01Anthropic priced Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, matching Opus 4.8 and undercutting Fable 5 by half.
- 02Opus 5 ships with stronger cybersecurity safeguards than Opus 4.8 after Fable 5's earlier forced takedown alongside Mythos 5.
- 03A new Fast mode launches in research preview at double the standard token price, with automatic fallback to a lower-tier model when safeguards decline a request.
- 04The release lands days after an OpenAI security incident and weeks after GPT-5.6 rolled out through a two-week government-only preview.
Asked whether Opus 5 went through government review before public rollout, Anthropic spokesperson Danielle Ghiglieri would not confirm a formal preview period of the kind GPT-5.6 received.
“We continue to work with our government partners to conduct their own independent testing of our models. This includes Opus 5.”— Danielle Ghiglieri, Anthropic spokesperson
The pricing structure is the other quiet story. Anthropic has faced criticism in recent months over allegedly hidden usage caps — the subject of at least one lawsuit — and broader industry unease about AI labs quietly passing rising compute costs to customers. Holding Opus 5 at Opus 4.8's price levels, and undercutting GPT-5.6 slightly, reads as a direct response. It also creates a clean two-tier pitch: Opus 5 at $5/$25 for most work, Fable 5 at double that for the frontier.
A new Fast mode launches today in research preview, offering higher throughput at double the standard token price. Customers can also opt in to automatic fallback to a lower-tier model when Opus 5's safeguards force it to decline a request — a feature aimed at enterprise developers who want continuity of service without engineering around every refusal.
The competitive frame matters. GPT-5.6 shipped under an unusual regulatory chaperone, and its rollout timeline was determined less by OpenAI than by federal reviewers. Anthropic's decision to release Opus 5 without an equivalent public preview restriction — while still emphasizing continued government testing — suggests the company believes its safeguard posture is strong enough to launch on its own timeline. Whether Washington agrees will show up in the next few weeks of usage.
The tighter safeguards do come with tradeoffs Anthropic has not detailed publicly. The optional lower-tier fallback exists precisely because Opus 5 will refuse more categories of request than Opus 4.8 did, and enterprise customers running production workflows will feel those refusals. The company has not published refusal-rate data comparing Opus 5 to its predecessor, and the coding benchmark improvements it cites have not been broken out with specific scores.
Opus 5's release confirms that the post-Fable 5 regulatory environment is now the baseline, not a one-off. Anthropic is threading a narrow lane — ship frequently, price flat, harden safeguards, and keep government partners in the loop — while OpenAI navigates the aftermath of its own security incident. For enterprise buyers, the practical result is more capable models at prices that are, for once, not climbing. That is a meaningful shift in a market where every prior frontier release has arrived with a higher sticker.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
