Is Claude Fable 5.2 worth the price?

On 10 November, Robbert van Empel published a video testing Claude Fable 5.2 by asking it to build a playable 3D South Park game out of Lego. Other models received the same assignment for The Future’s game benchmark. Fable scored 89 out of 100 and placed third out of 21 entries at the time of the test. Claude Opus 5.5, which tackled the same brief in September, scored 87.

Fable built a snowy planet with twelve recognizable characters, a road to Canada and a story that can be played through to the end. Players rebuild Kenny, collect pages from other characters and carry a golden brick to Stark’s Pond. They can also pick up, drop and throw bricks. The review found that the world felt sparse between landmarks and that most missions followed a fetch-and-deliver pattern. Reloading the game reset progress. The scores reflect how each game handled this particular brief across visuals, originality, interaction and execution. The Future’s South Park benchmark

The video also looks at how much a higher reasoning setting can add to the bill. Anthropic’s standard API rate for Fable 5.2 is $8 per million input tokens and $40 per million output tokens. In Artificial Analysis’s separate Intelligence Index, Fable scored 62 on High and 65 on Max. The average cost per task in that test rose from $9.08 to $26.54. These figures describe the Intelligence Index runs. The South Park entry has no recorded generation cost. High setting · Max setting

For developers choosing a model, the relevant comparison is what a higher reasoning setting costs and how much it improves the work they actually do. In this game test, Fable’s strengths were a complete story and an idea that made use of the spherical world. If a project depends more on building mechanics or replayability, those are features to test directly. Both Claude games are playable from the leaderboard.