OpenAI launched GPT-6 Sol on September 22, 2026, positioning the model against Anthropic's ecosystem with claims of higher scores on business workflow tasks at a fraction of the cost per task, according to a report by The New Stack. Anthropic released its competing Claude Opus 5.5 model the same day, prompting comparative trials that evaluate models beyond single-shot accuracy by measuring consistency across repeated runs.
Pricing and Initial Capability Claims
According to OpenAI's launch figures, GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, which is half the cost of GPT-5.6 Sol. Meanwhile, Anthropic's Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens. OpenAI marketing materials reported that Sol scores 33.2% on Zapier's AutomationBench compared to 26.9% for Opus 5. In initial single-shot tasks covering standard developer workflows—such as fixing bugs in a Python billing service, triaged CI jobs, and answering questions from a mock SDK—both GPT-6 Sol and Claude Opus 5.5 tied by scoring perfectly.
Multi-Run Consistency Testing
To separate models where accuracy ties, testing across five repeated API calls at maximum effort settings revealed operational differences. For CI log triage, both models achieved perfect scores across all five trials, with Sol averaging 18 seconds and $0.02 per run, coming in at approximately 8% of the cost of Opus 5.5. However, on harder tasks involving a 3,664-line incident log and a dependency resolver specification, Opus 5.5 achieved a perfect 15 out of 15 successful runs, while GPT-6 Sol scored perfectly on 12 of 15 runs, introducing errors such as missed customer refund details or syntax issues that caused import failures.
Cost, Speed, and Operational Trade-Offs
Across all 15 test runs, GPT-6 Sol totaled $2.68, which is roughly 16% of the $16.72 total cost incurred by Claude Opus 5.5, while also delivering faster completion times across every evaluation. The findings suggest that Sol suits high-volume tasks where human review or automated test suites verify outputs, whereas Opus 5.5 fits scenarios where incorrect answers carry high costs and human oversight is absent.
AI-generated editorial illustration; not a documentary photograph.
Source: GPT-6 Sol vs. Claude Opus 5.5: Is cheaper important when results aren’t consistent?
SIGNAL TITANS
Leave a Reply