Nace AI on 19 October 2026 released Drex Max, which it calls a multimodal decision model for computer use. In the benchmark table Nace published with the launch, a decision costs $0.06 per 1,000 with Drex Max, against $10.17 for Claude Opus 5.5 and $3.33 for GPT-6 Sol.
Nace says Drex Max is available today through its API at $0.06 per million input tokens, with a 262K-token context window. All the comparisons in this piece come from Nace’s own table.
Business Pill · MODEL ROUTING
A one-minute explainer of model routing. It teaches the general idea only and says nothing about Nace or Drex Max.
The key insight: As we read it, Nace is pricing each agent step as a decision rather than as written text, and that is where its own table shows the gap: $0.06 per 1,000 decisions for Drex Max against $10.17 for Claude Opus 5.5.
Nace describes Drex Max as “our most capable decision model, and the first one that sees.” A caller sends a screenshot, a screencast, a scanned document, a chart or a photo with a set of typed questions, and “one forward pass returns a probability for every option you supplied.”
The pitch is aimed at computer-use agents. Nace writes that a single task “can take thirty” small decisions, such as which button to click. It adds that each one today “usually means sending a screenshot to a large frontier model and waiting seconds while it writes an answer.”
Because the model writes no text, Nace says, “a decision costs one pass over the input no matter how many options it scores.” TestingCatalog, which flagged the launch on X, summed it up as: “Drex Max doesn’t write text; it replies only with one of the options it’s given.”
Cost per 1,000 decisions in Nace AI’s own benchmark table: Drex Max $0.06, GPT-6 Luna Decisions $0.10, GPT-6 Sol $3.33 and Claude Opus 5.5 $10.17. Rival figures were measured by Nace. About 170 times is our arithmetic ($10.17 divided by $0.06). Source: Nace AI blog (19 Oct 2026).
What Was Posted on X
Drex Max is here: our multimodal decision model text, images, and video 34 ms/decision, under 1s at 262K tokens $0.06 / 1M input tokens available today at https://nace.ai/drex
Nace AI announced Drex Max on X at 13:33 UTC on 19 October 2026, listing text, image and video input and a price of $0.06 per million input tokens. Its post gives 34 ms per decision; the median API latency in Nace’s blog table, network included, is 183 ms.
Nace AI has released Drex Max, its most powerful decision model to date, which can now read images, videos, and text.
Computer Use agents can send Drex Max a screenshot at each step, ask which action to take from a fixed set of options, and receive a probability score for each one.
> Drex Max doesn’t write text; it replies only with one of the options it’s given.
> Web agents can use it to select the right element of a page based on what is on screen.
> It runs up to 167ร cheaper per token than the models behind Claude, GPT, and Gemini for computer use.
— ๐จ AI News | TestingCatalog (@testingcatalog) Oct 19 14:17
TestingCatalog flagged the launch on X and wrote that Drex Max runs up to 167 times cheaper per token than the models behind Claude, GPT and Gemini for computer use. That figure is TestingCatalog’s; Nace’s blog table gives costs per 1,000 decisions, from which this piece computes about 170 times against Claude Opus 5.5.
The Numbers in Nace’s Table
On cost, Nace lists Drex Max at $0.06 per 1,000 decisions, GPT-6 Luna Decisions at $0.10, GPT-6 Sol at $3.33 and Claude Opus 5.5 at $10.17. By our arithmetic, that puts Drex Max at about 170 times below Opus 5.5 on this measure. Nace gives Drex Max a median API latency of 183 ms, against about 2.0 seconds for GPT-6 Sol and about 3.5 seconds for Claude Opus 5.5.
On OSWorld-Verified, a computer-use automation test in the same table, Drex Max scores 0.780, against 0.790 for Claude Opus 5.5 and 0.756 for GPT-6 Sol. On the public Decision Index in Nace’s table it scores 67.6, against 47.8 for GPT-6 Luna Decisions.
On general knowledge it trails the frontier models. Nace lists Drex Max at 72.5 on MMLU-Pro, against 93.0 for GPT-6 Sol and 94.0 for Claude Opus 5.5, and at 19.0 on HLE, against 39.0 and 56.0.
Connecting the Coverage
Drex Max joins a category we have been tracking. On 7 October we reported that OpenAIpriced its Decisions API, which runs on GPT-6 Luna and returns typed answers rather than generated text, at $0.10 per million input tokens with no output charges. Nace’s table lists that product as GPT-6 Luna Decisions.
On 20 October we covered Cloudflare’s Clairic, which Cloudflare calls an extraction model and places alongside its own decision model, Clef. Cloudflare credits the idea to TypeSafe and its decision model, Jev, a version of which also appears in Nace’s table.
The Structural Read
As we read it, the table splits an agent’s work in two: the frontier models still lead on general knowledge, at 93.0 and 94.0 on MMLU-Pro against 72.5 for Drex Max, while the per-step choices are where Nace prices its model.
As we read it, Nace places Drex Max in the same category as OpenAI’s Decisions API, which its table lists as GPT-6 Luna Decisions at $0.10 per 1,000 decisions.
As we read it, speed is the other half of the pitch: Nace lists a median of 183 ms per decision for Drex Max, against about 3.5 seconds for Claude Opus 5.5.
Nace AI blog, 19 October 2026
“A generative model pays for every token it writes. Drex Max writes nothing, so a decision costs one pass over the input no matter how many options it scores”
Three Implications
PRICED PER DECISION Nace lists Drex Max at $0.06 per million input tokens, and at $0.06 per 1,000 decisions in its table.
BENCHMARKS ARE NACE’S Every rival figure in the table was measured by Nace, not by an independent lab.
A GROWING CATEGORYOpenAI’s Decisions API returns typed answers rather than generated text, and Cloudflare sells its own decision model, Clef, as our earlier pieces reported.
The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, Drex Max is aimed at that marginal cost: Nace says a decision costs one pass over the input no matter how many options it scores. The framework was published in March 2026 and does not discuss Drex Max.
What Is Not Established
Every figure here comes from Nace’s own launch post, including its measurements of rival models. We have not seen an independent benchmark of Drex Max, and the post does not explain how it calculated the cost per 1,000 decisions for rival models.
Business Engineer Framework
The State of The Inference Economy
A Business Engineer framework on why inference, not training, now drives the cost and revenue of AI.
Nace is selling Drex Max as the model an agent calls for each small choice, priced at $0.06 per 1,000 decisions in its own table, while frontier models in the same table cost far more per decision and score higher on general knowledge.
A note on sourcing. This piece is built on Nace AI’s launch post of 19 October 2026, read in full, including its benchmark table, plus TestingCatalog’s post on X and our own earlier pieces on decision models. Every benchmark and cost figure is Nace’s, including its measurements of rival models; we have not run the model. The about-170-times figure is our arithmetic. Nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
