The AI model market continues to show paradoxical dynamics: Anthropic's newest model, Claude Fable 5, confidently won a test simulating realistic physics, but the price of this superiority turned out to be astronomical. According to my analysis of data from the atomic.chat service, Fable 5 outperformed competitors in the task of creating three independent HTML5 scenes demonstrating physics, yet its operating cost was roughly six times higher than its predecessor, Claude Opus 4.8.
Test Details: From a Collapsed Bridge to a Monster Truck
Four leading models — Fable 5, GPT-5.5, Opus 4.8, and GLM-5.2 — were tasked with simulating three episodes: a train derailing from a collapsed bridge into water, two cars colliding mid-air after a ramp jump over a canyon, and a scene with a monster truck crushing a row of parked cars. The key requirement was realistic physics of falling, destruction, and collisions without extraneous artifacts.
Fable 5 handled the task flawlessly, achieving the maximum score across all three scenes. Collisions looked realistic, objects fell and broke correctly, without clipping through the ground or hanging in mid-air. However, this quality came at a cost: the model consumed 62,158 tokens, with a price of $3.12.
Comparative Analysis: Price vs. Quality
The closest competitor to Fable 5 was GPT-5.5, which consumed 37,753 tokens at a cost of $1.14. In the monster truck scene, it even performed slightly better than its rival. However, the most interesting contrast is shown by Opus 4.8: with only 22,280 tokens consumed and a cost of $0.56, it was six times cheaper than Fable 5, though the quality of its scene rendering was noticeably more modest.
The most economical model was GLM-5.2: it didn't win any scenes, but cost only $0.08 with 36,246 tokens. This is a classic example of the trade-off between quality and cost.
Accessibility Context: Return After Restrictions
It's important to note that the test was conducted shortly after Fable 5 returned to full availability. Recall that on June 30, U.S. authorities lifted export restrictions on Anthropic's Claude Fable 5 and Mythos 5 models, with global access restoration beginning on July 1. The restrictions were imposed on June 12, just three days after the models' release, citing national security concerns. Anthropic was forced to disable both models for clients on Claude.ai, via API, AWS Bedrock, and other partner services, as real-time filtering of users by nationality proved impossible.
Fable 5 is positioned as Anthropic's most advanced and widely available model. It is built on the Mythos-class architecture with added finely tuned safety mechanisms and excels at complex tasks, programming, and image analysis. Internal classifiers automatically redirect high-risk requests to safer algorithms.
My expert conclusion: Fable 5's victory in the physics test is an undeniable achievement, but the sixfold price difference compared to Opus 4.8 raises questions about its practical viability for mass use. For tasks where every detail of physical simulation is critical, Fable 5 is justified, but for routine scenarios, Opus 4.8 remains a more rational choice. The AI market is increasingly splitting into niche premium solutions and affordable universal tools, and this test is a vivid illustration of that trend.