The frontier of artificial intelligence has rarely moved this fast. Within a few weeks, Anthropic has released two models at the top of its range, OpenAI has previewed a new three-tier series, and a Chinese laboratory has published an open model that unsettled financial markets. For anyone trying to keep track, here is where the leading systems now stand — and what actually separates them.
The models at a glance
| Model | Developer | Positioning | Access & restrictions | What stands out |
|---|---|---|---|---|
| Claude Opus 5 | Anthropic | Heavyweight general model, positioned as the practical workhorse | Smaller than Fable 5, cheaper and less restrictive; not covered by the 30-day data retention policy that applies to Fable and Mythos | Anthropic says it comes close to Fable 5 in many domains and is much stronger at complex coding and at verifying its own work |
| Claude Fable 5 | Anthropic | Mythos-class flagship, the most capable tier | Subject to a 30-day data retention policy; was taken offline for several weeks alongside Mythos 5 before returning with stronger cyber safeguards | Its capabilities triggered a standoff with the US government that opened a new phase of AI regulation |
| Kimi K3 | Moonshot AI (China) | Open model presented as a rival to leading US labs | Released as an open model, unlike the closed US flagships | Went viral and rattled financial markets; the White House alleged it was distilled from Anthropic's Fable, a claim several experts dispute |
| GPT-5.6 (Sol, Terra, Luna) | OpenAI | Three-tier series: Sol as flagship, Terra balanced, Luna low-cost | Introduced through a limited preview, with general availability announced for the following weeks | OpenAI says Terra matches GPT-5.5 performance at half the cost, and that the series ships with its most robust safety stack to date |
Anthropic ships Opus 5 — and makes it the practical choice
Anthropic released Claude Opus 5 this week, only two months after Opus 4.8. The company says the model "comes close to the capabilities of Claude Fable 5 in many domains" and is considerably stronger at complex coding tasks, describing it as "much stronger at verifying its work and iterating carefully until it succeeds".
The more consequential difference may be commercial rather than technical. Opus 5 is smaller, cheaper and less restricted than Fable 5, and it is not covered by the 30-day data retention requirement that applies to Anthropic's most powerful tier. For many organisations, that combination makes it the more usable of the two — even where Fable 5 remains nominally more capable. Anthropic also states that Opus 5 outperforms Fable 5 on several of the benchmarks included in its announcement.
Fable 5 and the regulatory turn
Fable 5, the public-facing model in Anthropic's Mythos class, has had an eventful few months. Its capabilities drew scrutiny from the US government, and the model was taken offline for several weeks alongside Mythos 5 before returning with reinforced cyber safeguards. Those negotiations are widely seen as the opening of a new phase of AI regulation in Washington — context that helps explain why Opus 5 launched with stronger cyber protections than the Opus generation before it.
Kimi K3: the open challenger
Moonshot AI, a Chinese laboratory, unveiled Kimi K3 in mid-July, claiming it can compete with the leading American systems. What made it a market event was less the model than the reaction to it: the release went viral and, in the words of one industry account, left Wall Street rattled about the competitive position of US labs.