Gemini 4 Carbon vs GPT-6.1 Sol: Pricing and Performance

At a glance

Gemini 4 Carbon and GPT-6.1 Sol are both frontier reasoning models, but they are aimed at different buyers right now. Carbon is Google’s new top-ranked model with a strong focus on cybersecurity and long tasks, and most teams cannot use it yet. GPT-6.1 Sol is OpenAI’s cost-focused frontier model, available today through the API.

Attribute	Gemini 4 Carbon	GPT-6.1 Sol
Maker	Google DeepMind	OpenAI
Release	October 28, 2026	September 29, 2026
Access	Cyber defenders and Fairwind Program partners first; wider access “coming soon”	OpenAI API
Input price (per 1M tokens)	$2 introductory, $4 later	$2
Output price (per 1M tokens)	$10 introductory, $20 later	$10
Cached input	95% off the input price	$0.10 per 1M tokens
Best for	Security work, long multi-step engineering tasks	General reasoning at a lower cost

Gemini 4 Carbon: strengths and limits

Google positions Carbon for long agentic work, saying internal teams saw it work through dead ends over days on hard engineering problems, such as tracing subtle bugs across large codebases.

Its security results stand out. Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them, and 71.0% on CWE-bench for patching. It says the Chrome team found Carbon more than 10 times better at generating patches than another lab’s frontier model, and that internal secure-code tests showed more than twice the success rate of Gemini 4 Argon.

Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, according to Google, and placed fourth on WebDev Arena at 1768 Elo. On the SemiAnalysis GDPx benchmark of 206 work tasks, it was reported to finish close behind Claude Opus 5.5 at about $16 per task against Opus’s $51. When SemiAnalysis left out the cybersecurity, biology and alignment tasks, GPT-6 Astra and GPT-6.1 Sol ranked above Carbon, which suggests its lead is mostly in security work.

The main limit is access. Google is releasing Carbon to cyber defenders first, then to trusted partners through its Fairwind Program, and has not given a date for general availability. The low price is also temporary: after the introductory period, Google lists $4 per million input tokens and $20 per million output tokens, double the launch price.

GPT-6.1 Sol: strengths and limits

GPT-6.1 Sol is OpenAI’s cost-focused frontier model. It launched on September 29, 2026, a week after GPT-6 Sol, and halved the cache read price to $0.10 per million tokens. OpenAI reports that it costs $5.47 per task on average at maximum effort on its own tests, against $23.21 for Claude Opus 5.5.

Its strength is breadth at a lower price. On the SemiAnalysis GDPx results with security and biology tasks removed, GPT-6.1 Sol ranked in the top group alongside Claude Sonnet 5.5. For everyday reasoning, writing and coding work, that combination of a good score and a low price is the main reason to pick it.

The limits are on the specialist side. OpenAI does not pitch Sol as a security model, and in our GPT-6.1 Sol vs Claude Opus 5.5 comparison, Opus 5.5 scored 66.4% on Terminal-Bench 4.0 against Sol’s 37.3%. There are no published tests that run Sol and Carbon on exactly the same tasks.

Price and access

During Carbon’s introductory period, both models cost the same per token: $2 per million input tokens and $10 per million output tokens. After that, Carbon costs twice as much as Sol per token. Caching works differently: Google takes 95% off Carbon’s input price for cached tokens, and OpenAI charges $0.10 per million cached tokens for Sol.

Be careful with per-task costs. The $5.47 figure for Sol comes from OpenAI’s own tests, and the $16 figure for Carbon comes from the SemiAnalysis benchmark. They measure different tasks, so they cannot be compared directly. The only fair cost comparison is one you run yourself on the same work with both models.

Sol is available now through OpenAI’s API. Carbon is limited to approved security teams and partners for now. Prices can change, so check both providers’ pricing pages before you budget.

Which one should you choose

Choose Gemini 4 Carbon if you work in security, such as finding and patching vulnerabilities, and you have access through Google’s programs. Its published security results are strong, though they are reported by Google and not yet independently checked.

Choose GPT-6.1 Sol if you need a frontier-level model today for general work and want predictable per-token costs. For most teams, the choice is simple for now: Sol is available, and Carbon is not.

If you are planning ahead, budget Carbon at the later $4 and $20 price, not the launch price, and re-test when Google opens general access.

Common questions

Is GPT-6.1 Sol better than Gemini 4 Carbon?

It depends on the work. Carbon ranks first on the Artificial Analysis index and leads on security benchmarks. When SemiAnalysis removed security and biology tasks from its benchmark, GPT-6.1 Sol ranked above Carbon. For security work Carbon looks stronger; for general work Sol is competitive and available now.

Is GPT-6.1 Sol cheaper than Gemini 4 Carbon?

Per token, they cost the same during Carbon’s introductory period, $2 input and $10 output per million tokens. After that, Carbon’s price doubles to $4 and $20, so Sol becomes half the price per token. Published per-task costs come from different benchmarks and should not be compared directly.

When will Gemini 4 Carbon be widely available?

Google has not given a date. Carbon is available first to cyber defenders, then to trusted partners through the Fairwind Program, with wider access for developers, businesses and consumers described as coming soon.

The conclusion

Gemini 4 Carbon is the stronger model on paper for security and long engineering tasks, but it is not yet available to most teams, and its price will double after the launch offer. GPT-6.1 Sol is the practical choice today: available through the API, competitive on general work, and cheaper per token once Carbon’s introductory price ends.

Watch for three things: a general availability date for Carbon, independent tests that run both models on the same tasks, and how long Google keeps the introductory price.

For a closer look at each model, see our Gemini 4 Carbon launch analysis and the GPT-6.1 Sol and Claude Opus 5.5 price cuts.