I Tested GPT 5.6 Sol vs Fable 5. What You Need To Know.

AI summaries can be incomplete or wrong. Verify anything important against the original video.

This video compares GPT-4.5 Sol and Claude Fable 5 models on various benchmarks, finding Sol to be more cost-effective and faster, while Fable excels in creativity and certain complex reasoning tasks. The reviewer highlights Sol's lower cost, faster API, and better performance on benchmarks, while Fable offers more nuanced capabilities but at a higher price and slower speed. The video concludes with a verdict favoring Sol for general use and Fable for specific creative or complex tasks.

• Cost-effectiveness

  Sol is significantly more cost-effective than Fable, costing roughly half as much for comparable API calls and tasks.

  • 15:01 API benchmarks show Sol costs $9.01 vs Fable's $14.72 for 130 calls.
  • 11:46 Agentic builds show Sol costs $7.06 vs Fable's $48.68.
  • Total cost for the session: Sol $16.07 vs Fable $63.40.
• Speed and Latency

  Sol is faster than Fable in terms of median API latency and mean API latency.

  • Median API latency: Sol 8.2s vs Fable 10.0s.
  • Mean API latency: Sol 34.1s vs Fable 23.4s (this seems reversed in the video, review might be wrong).
• Reliability and Performance

  Sol has superior reliability with 0 non-answers compared to Fable's 29 non-answers.

  • Sol: 0/130 never refused, never timed out, never truncated.
  • Fable: 29/130 runs produced nothing: silent classifier refusals (unpredictable domains, no explanation), one timeout, spiral, thinking-token truncations. Everything else about it is competitive.
• Quality and Capability

  While Sol excels in cost and reliability, Fable demonstrates strengths in creativity and complex reasoning, making it potentially better for specific tasks.

  • Fable's entire quality gap is non-delivery, not wrong answers. Answers all came on the hardest work in the pool: the most complex engineering build twice (a mini-spreadsheet engine, a DST-correct event expander) and a judgment-heavy policy-cases task. Sol's rare misses were self-inflicted slips (a garbled hand computation, one syntax error, one operator-precedence bug).

• 0:50 Overview of the comparison methodology, including the types of benchmarks and tasks used.
• 4:10 Analysis of cost, with Sol being significantly cheaper than Fable across various metrics.
• Introduction to GPT-4.5 Sol and Claude Fable 5 as the models being compared.
• Comparison of performance and speed, highlighting Sol's faster API response times.
• Discussion of quality and reliability, noting Fable's strengths in creativity and nuanced reasoning, while Sol excels in direct answers and fewer errors.
• Reviewer's verdict: Sol is better for general use, Fable for specific creative or complex tasks.