Gemini 4 Carbon Benchmark Results for Data Analysis

Nov 10, 2026: Google's strongest model, tested on real-world data analysis.

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.

On October 28, Google announced Gemini 4 Carbon, which it calls its strongest model, built for long-horizon coding, agentic work and cybersecurity. Google first gave access to a small group of cyber defenders, and this week Carbon reached developers through Google AI Studio and OpenRouter. It has a context window of 1 million tokens. Google says Carbon builds on Gemini 4 Argon, released two weeks earlier, but hasn't said how it was trained or how many parameters it has, and the weights aren't public. It launched at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens, with cached input 95% off.

We ran it through our data analytics exam. Gemini 4 Carbon passed 42 of 43 questions (98%) for $11.55, one point better than Gemini 4 Argon and a little cheaper. It found all five hidden patterns and caught all five statistics traps. That ties it for fourth on our board with Grok 4.7, Kimi K3.1 Pro and Tencent's Hy4. But it's slow, with a median of 93 seconds a question, and Hy4 got the same score for $2.11. On the chart below, Carbon sits off the frontier.

Gemini 4 Carbon vs. Gemini 4 Argon, GPT-6.1 Sol and Hy4

*1,000 people each asking 10 questions a day, 250 working days a year, at OpenRouter's prices when each model ran.

Carbon was perfect everywhere except data quality, where it got two of the three planted problems, the same as Argon. It also fixed the one analytics question Argon got wrong. Only three models have done better: GPT-6.1 Sol, GPT-6 Astra and Claude Fable 5.2, which all scored 43.

One run per model is a single sample. In our repeat runs of other models, scores moved by up to three points between runs, so treat a one- or two-point gap as close to a tie.

Thorough and slow

Carbon takes its time. Its median question took 93 seconds, four times as long as Argon's 23, and only five of the 46 models on our board are slower. Its whole exam took about 1 hour 48 minutes. On the five questions with hidden patterns it ran 41 queries on average, against a board median of about 25, and it found all five.

That work shows up in tokens, but caching keeps the bill down. Carbon read about 29.3 million input tokens, and 88% of them came from the cache at a twentieth of the input price. Our run cost $11.55, a little less than Argon's $13.47. Google calls the current price introductory, so expect that number to rise.

Where it got things wrong

Carbon missed one question. On a data quality audit, the first problem it reported wasn't a real defect: it read an expected feature of the data as missing data, and put that ahead of the real problems it found. Its explanation was careful and its other findings were right, but an analyst would have sent the team chasing a problem that doesn't exist.

The Test Harness

We run every model in Plotly Studio Embedded, a drop-in analytics agent that runs on Dash apps. This harness is tuned to run data analytics queries and make charts with multi-turn tool calling, with less emphasis on writing text. It's meant to put the data front and center.

Frequently asked questions

How good is Gemini 4 Carbon at data analysis?

On PlotlyDataAnalyticsBench it passed 42 of 43 questions, tied for fourth with Grok 4.7, Kimi K3.1 Pro and Hy4. It found every hidden pattern and caught every statistics trap. GPT-6.1 Sol, GPT-6 Astra and Claude Fable 5.2 scored 43.

Is Gemini 4 Carbon better than Gemini 4 Argon?

On our exam, by one point: 42 against Argon's 41. It cost a little less per run, but it was four times slower.

How much does Gemini 4 Carbon cost?

Google's introductory price is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. Our whole exam cost $11.55. GPT-6.1 Sol scored 43 for $2.61.

How fast is Gemini 4 Carbon?

Slow. Its median question took 93 seconds, against 18 for GPT-6.1 Sol and 23 for Gemini 4 Argon.

Is Gemini 4 Carbon open source?

No. Google hasn't released the weights. You can use it through Google AI Studio and providers such as OpenRouter.

About the benchmark

We run new models through the same exam as they're released, and we'll keep sharing what we find. We ran Gemini 4 Carbon through OpenRouter with the same harness and grading as every other model. Google AI Studio, its only OpenRouter provider, caps each response at 32,768 tokens, so we capped our requests to match.

To see all 46 models on one chart and in a full results table, along with cost, speed and how the exam works, see Introducing DataAnalyticsBench.

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.