Claude Model Lineage 2026: From Opus 4.5 to Fable 5

Seven releases, roughly one every four weeks. The Opus 4.x line was an iterative grind: bigger context, agent orchestration, sharper vision, better self-checking, each shipped without moving the flagship price. Then the generation flipped. Fable 5 arrived as the public face of a deliberately split release, with its restricted twin Mythos 5 reserved for vetted cyberdefenders. That split is the clearest signal in the whole timeline that capability had reached a point where Anthropic no longer wanted to ship the full model to everyone at once.

For day-to-day work the practical read is simple. Opus 4.8 remains the workhorse: $5/$25, strong coding, and the model that Fable 5 quietly routes sensitive requests back to. Fable 5 is the upgrade you reach for when a job is genuinely long-horizon or agentic and the extra capability earns its doubled price. Sonnet 4.6 stays the value pick when you want near-Opus quality at $3/$15. Everything else in the list is history that explains how those three options came to exist.

Same data, one table. The price column makes the flat-then-double pattern obvious, and the SWE-bench column reads as a clean climb with one flat step at Opus 4.6. SWE-bench Verified figures are Anthropic-reported.

Strip away the feature announcements and a handful of figures carry the trend. SWE-bench Verified is the cleanest of them because Anthropic reported it consistently across the line, which makes the climb directly comparable release to release. Treat all four below as Anthropic-reported except the independent Harvey figure, which we have flagged.

The honest caveat: a single benchmark is a proxy, not a verdict. SWE-bench Verified tracks one slice of coding ability, and the jump from Opus 4.8's 88.6% to Fable 5's 95.5% is real but does not mean every task got 7 points easier. Use it as a directional signal, then test on your own workload before you commit a pipeline to the newer, pricier model.

Until April 7, 2026 the lineage was a single ladder of Opus releases. Mythos Preview broke that pattern. Anthropic introduced it as a new capability tier sitting above Opus, and the reason it was restricted rather than sold tells you what changed: the model showed a step change in cybersecurity ability, including autonomous discovery and exploitation of previously unknown vulnerabilities. That is dual-use power, and Anthropic kept it inside Project Glasswing, available only to vetted partners under restriction.

Because it was never generally available, Mythos Preview carried no public API price, and we are deliberately not putting a number on it. The one published figure we can cite is 64.7% on Humanity's Last Exam with tools, an Anthropic-reported result. The class continued into the Claude 5 generation as Mythos 5, the restricted twin of Fable 5: the same underlying model, with safeguards lifted in some areas for cyberdefenders rather than routed away from risk.

For most builders the Mythos line is context rather than a product you will touch. It matters because it explains the split-release design at the end of the timeline. Once a model is capable enough that its full abilities are genuinely risky, shipping one version to everyone stops being the default. If you want the deeper background, see what Claude Mythos is.

The lineage rewards a boring, deliberate posture: pin to a model, measure it on your own work, and only chase the newest release when a specific job needs it. Here is how the three live options map to real decisions.

If you are weighing the latest jump specifically, the head-to-head detail lives in Fable 5 vs Opus 4.8. For the broader question of where Claude fits against other vendors, the comparison shelf in the Anthropic Claude hub is the place to start.