OpenAI has released three new separate models: Sol, Terra, and Luna. The competition now centers on Sol versus Anthropic's Claude Fable 5.
Sol costs $5 per million input tokens and $30 output. Fable 5 is $10 and $50. Anthropic has extended free access to Fable 5 for paid subscribers through July 19, following a U.S. government ban in June.
Benchmarks favor Sol. On the Artificial Analysis Coding Agent Index, Sol scored 80 to Fable's 77.2. On the Agents' Last Exam, Sol hit 53.6% against Fable's 40.5%. On Terminal-Bench 2.1, Sol reached 91.9% to Fable's 83.1%.
In creative writing tests, Fable produced more culturally specific narratives with cleaner causal logic. In an associative thinking test, the models tied, with Sol offering more explicit explanations and Fable relying on implicit metaphor.
In a logic puzzle, both models gave the same incorrect answer, failing to notice a missing constraint in the prompt. Both defaulted to a cached solution.
For a coding test, Fable 5 built a more polished game with music and animation. Sol produced a functional but visually simpler version.
The choice depends on use case. Sol offers superior performance on coding benchmarks at a lower cost. Fable 5 shows an edge in nuanced creative tasks. Anthropic's deadline extensions and pricing may influence the final decision.