We're officially recommending an OpenAI model as the default model family for Omni customers

Back to Devoured - October 01, 2026

DEVOURED

We're officially recommending an OpenAI model as the default model family for Omni customers

Omni benchmarked 16 model configurations and found that OpenAI's GPT-6 Sol outperforms more expensive models by prioritizing high-frequency reasoning over raw intelligence.

What: Marc Dupuis, founder of Omni, reported that GPT-6 Sol achieved 92% accuracy on their internal analytics evaluation at a cost of $0.23 per question, outperforming Anthropic's Claude 3.5 Sonnet and Opus models, which scored 72% ($0.19) and 87% ($0.54) respectively.

Why it matters: This indicates that for specialized analytical tasks, models that utilize frequent, iterative reasoning steps—even if they cost more per call—can provide higher reliability and better value than 'smarter' models that attempt to resolve problems in fewer passes.

Deep dive

Omni tested 16 combinations of OpenAI and Anthropic models against a proprietary analytics benchmark.
GPT-6 Sol reached 92% accuracy compared to 87% for Claude Opus 3.5 and 72% for Claude Sonnet 3.5.
The performance shift is attributed to 'persistence'—making more model calls per question rather than relying on a single, complex pass.
Cost efficiency was measured by the total cost to complete an answer, rather than price per token.
The model requires more total model calls per output compared to previous defaults, but significantly reduces spending on prompt caching.
Latency for GPT-6 Sol was recorded at 29 seconds median versus 40 seconds for the competition.

Decoder

Prompt Cache: A feature that stores frequently used tokens or system instructions to reduce latency and costs in repetitive AI workflows.
Evaluation (Eval): A specialized dataset and testing framework used to measure an AI model's performance on specific, measurable tasks like data analysis or coding.

Original article

We're officially recommending an OpenAI model as the default model family for Omni customers. Before making the call, we ran GPT-6 Sol against Claude Sonnet 5.5, Sonnet 5, Opus 5.5, and Opus 5 on our hardest analytics eval. Sol got 92% right at $0.23 a question. Sonnet 5.5, out this week, got 72% at $0.19. Sonnet 5, our default until now, got 60% at $0.32. Opus 5.5 came closest at 87%, but it cost $0.54 a question. (Sol was faster too, 29 seconds median vs 40.) It's interesting where the savings came from. Sol made almost twice as many model calls per question as Opus, about nine to five. But Opus spent about half its bill writing to the prompt cache, and Sol wrote a lot less. We ran 16 model and effort combinations across OpenAI and Anthropic. Sol on medium effort came out on top. Customers have been asking us for new models, and since last week I've been pointing them at Sol. So for now, new US customers start on GPT-6 Sol. If Claude wins the next round, we'll switch back.

Explore all 58 articles in today's digest