Model ML completes finance work more efficiently with GPT-5.6 Sol
The finance AI startup is running research, analysis, and deliverables through OpenAI's newest model — and cutting the rework that usually follows.
Written by OutOfToken AI
August 10, 2026 · 4 min read · Synthesized from reporting by OpenAI Blog · How this works
Finance teams have long treated AI-generated slides and spreadsheets as rough drafts at best — something to be checked, reformatted, and often rebuilt by hand. Model ML says that's changing with GPT-5.6 Sol, the model it's using to take finance work from raw research to client-ready PowerPoint decks and Excel workbooks, with fewer errors and less cleanup along the way.
From research to deliverable, in one pass
Model ML builds AI tools for finance professionals, and its workflows span stock analysis, regulatory interpretation, and the kind of client deliverables that traditionally eat up analyst hours. The company's pitch for GPT-5.6 Sol isn't just smarter answers — it's outputs that survive contact with a real client meeting without needing a rebuild first.
Fewer tokens, fewer mistakes
OpenAI says GPT-5.6 Sol generated decks using 39% fewer tokens per deck than Anthropic's Claude Fable 5, based on tests across 20 client workflows and hundreds of decks run through Model ML's internal FinBench evaluation. Crucially, the token savings didn't come at the cost of quality — OpenAI reports the decks were more polished, with clearer and more accurate data visualizations, and needed less rework before they were shared with clients.
""GPT‑5.6 Sol is the first model we've evaluated that consistently generates decks ready for real work," said Chaz Englander, co-founder and CEO of Model ML."
Why token efficiency actually matters here
In finance workflows, token count isn't just a cost line item — it correlates with how tightly a model reasons. A model that pads its output with unnecessary detail is more prone to inconsistencies across a 40-slide deck or a multi-tab workbook, and someone still has to catch those before the material goes to a client. Separate OpenAI evaluations covering financial, medical, and legal prompts found responses from the updated Sol model contained at least one factual error about 68% less often than GPT-5.5 Instant, a meaningful signal for a domain where a single wrong figure can undermine an entire report.
Traceability as the real unlock
The bigger shift for finance teams may not be raw accuracy but traceability — outputs where the underlying logic, sourcing, and calculations can be audited rather than taken on faith. Model ML's framing of decks as "ready for real work" implies exactly that: analysts can edit and verify a Sol-generated file rather than treating it as a black-box draft that needs reverse-engineering before anyone trusts it.
GPT-5.6 Sol arrived as part of OpenAI's broader GPT-5.6 family, released July 9, 2026 alongside Terra and Luna, in a market where flagship context windows now regularly exceed 1 million tokens and per-token pricing keeps compressing. Whether other finance-focused startups follow Model ML's lead will likely hinge on whether these token and error-rate gains hold up across messier, real-world workloads rather than curated benchmarks — but for now, Model ML is treating it as the first model that can go from research to a finished deliverable without an analyst redoing the last mile.
Editorial Note
All major factual claims in this article—including the 39% token efficiency gain, the 68% error reduction rate, Model ML's business focus, and the quality improvements—are corroborated by the provided research sources, particularly OpenAI's official announcement and third-party benchmarking articles. The research confirms the benchmarking methodology and specific performance comparisons cited.
Claim Tracker
AI-assessed
Source 2 (OpenAI official) and Source 1 (AiMultiple benchmark article) both cite this exact 39% figure and the 20 workflows benchmark from Model ML's FinBench evaluation.
Source 3 (Visual Studio Magazine) confirms this exact finding: 'Responses containing at least one factual error were about 68% less common with the updated Sol than with GPT-5.5 Instant, according to OpenAI.'
Source 1 (AiMultiple) confirms LLMs assist in 'stock analysis, regulatory compliance' and Source 2 directly quotes Model ML CEO, validating the company's stated focus areas.
Source 2 (OpenAI official page) includes direct quote from Model ML CEO stating decks were 'more polished, legible decks with clearer, more accurate data visualizations that required less rework.'
Source 2 (OpenAI official) identifies 'Chaz Englander, Co-Founder & CEO at Model ML,' confirming both his role and title.
Ask AI about this story
// discussion
sign in to join the discussion