Gemini 4 Argon Benchmark Results for Data Analysis

Oct 13, 2026: Gemini 4 Argon scores an A on our data analytics exam, and is only the third model to find every pattern we hid in the data.

At Plotly, we keep a close eye on model releases. Our agentic analytics products, Plotly Studio Desktop and Plotly Studio Embedded, and our open source libraries Plotly and Dash work with any model, and our customers look to us for recommendations. Today Google released Gemini 4 Argon. It costs $2 per million input tokens and $10 per million output tokens, the same list price as GPT-6.1 Sol and Claude Sonnet 5.5.

We ran it through our data analytics exam. It passed 41 of 43 questions (95%, an A), level with Claude Opus 5.5, Muse Wonder, Fugu Ultra v2 and GPT-6 Sol, behind only GPT-6.1 Sol and GPT-6 Astra (43 of 43) and Grok 4.7 (42). It found all five patterns we hid in the data, which only GPT-6.1 Sol and GPT-6 Astra had done before.

The exam

Our data analytics exam tests how well models do at real data analysis. It runs within Plotly Studio Embedded, with the same agent, tools and prompts that app viewers use and only the model swapped, and it answers 43 questions through a series of tool calls and analysis steps. The data warehouse is representative of the real world: 20M rows across 44 tables and 554 columns. The questions cover a broad range of data analytic skills: answering questions, recognizing ambiguity, avoiding hallucinations, and finding a needle in a haystack.

The dataset is a mock wind turbine analytics dataset: 5,200 turbines across 28 wind farms over two years. It covers everything from hourly SCADA telemetry and alarms to work orders, inspections, oil samples, firmware rollouts and weather. And like real operational data it's messy, with frozen sensors, duplicated rows and turbines decommissioned mid-period, so a good analyst has to catch problems before quoting a number.

Every question has either an answer, checked against a verified key, or "a catch": an ambiguous question to raise, an impossible request to decline, a planted data defect to flag. There is no partial credit, and a confident answer that ignores the catch fails.

Here's how Gemini 4 Argon compares with Claude Opus 5.5, which has the same score and almost the same cost, and with GPT-6.1 Sol, which has the same price per token:

Cost and speed

It cost $13.47 in inference tokens for Gemini 4 Argon to complete the exam, about 31 cents per question, almost exactly what Claude Opus 5.5 cost ($13.31).

Gemini 4 Argon has the same price per token as GPT-6.1 Sol, but cost five times as much for the exam. The difference is how much it reads. Gemini 4 Argon went through about 510,000 tokens per question, against 183,000 for GPT-6.1 Sol and 215,000 for Claude Opus 5.5. Cheaper or faster tokens don't always mean cheaper or faster analysis, and the same price per token doesn't mean the same price per question.

For a sense of scale: if 1,000 people each asked it 10 questions a day, at the cost per question we measured, it would come to about $780 per person per year. GPT-6.1 Sol would cost about $152.

It answers a typical question in 23 seconds, a little faster than Claude Opus 5.5 and Muse Wonder. Its average is higher, 63 seconds, because it spends much longer on the hardest questions.

Where it got things right

Gemini 4 Argon did best where most models struggle:

• Hidden needles (5 of 5). We planted five patterns in the data for the model to find on its own, without being told to look. Gemini 4 Argon found every one, with its own numbers to back each up.
• Statistics traps (5 of 5). Asked whether windy days drive up SCADA alarms, it broke the alarms down by type instead of answering from one overall correlation. Seven of the 26 models on our leaderboard got all five statistics questions right.
• Ambiguous questions and impossible requests (10 of 10). It said which reading it used when a question had more than one, and said so when the data couldn't answer.

On the request with the warning "I'll be quoting these numbers in the monthly ops report, so make sure they're right", it found the duplicated rows, gave the corrected averages and flagged the missing last day of the month.

Where it got things wrong

Gemini 4 Argon missed 2 of the 43 questions:

• Declining a hard query. Asked for the turbines with the most hours under at least one active alarm, merging overlapping alarms, it said "I cannot write a verifiable query to merge arbitrary overlapping intervals" and reported unmerged alarm hours instead. That gives a different top five. Every other model that scored 40 or more got this question right.
• Claiming full coverage. Summarizing a quarter of gearbox oil temperatures, it described every turbine as having full hourly telemetry and missed a week of missing readings.

About the benchmark

We run new models through the same exam as they're released, and we'll keep sharing what we find.

The exam is private and we're sharing it with researchers. We're keeping it private so that the answer key doesn't make its way into the training set. For the same reason, the data is fabricated: the warehouse is a synthetic wind-energy fleet, built to look like a real operator's data, so its answers don't exist anywhere else.

If you'd like access, reach out to Chris Parmer at Plotly.