Gemini 4 Carbon vs Gemini 4 Argon

Use Gemini 4 Argon for production workloads today because Google has not opened public access to Gemini 4 Carbon. At Layer3Labs, we build launch guides for new artificial intelligence (AI) models and route client workloads across vendors, so we check which models developers can actually call before we compare scores.

Google announced Gemini 4 Carbon on October 28, 2026, as its strongest frontier model. Gemini 4 Carbon is restricted to a subset of cyber defenders through the Fairwind Program, and Google lists broader developer access only as coming soon.

Argon runs on the Gemini API today with the model identifier gemini-4-argon at introductory rates of $2 per 1M input tokens and $10 per 1M output tokens. Gemini 4 Carbon has the same introductory rates, so waiting for Carbon does not lower the price per token.

Gemini 4 Carbon vs. Gemini 4 Argon: Side-by-Side

Are you one of these vendors? Update your listing

Use Argon Now, Switch When Carbon Opens

Gemini 4 Argon is the only operational choice between the two models for live software environments as of October 28, 2026. Google launched Gemini 4 Argon about two weeks before Carbon, and Argon is generally available on the Gemini API outside the free tier.

Gemini 4 Carbon represents a clear capability upgrade across all 17 benchmarks Google published, but access is confined to partners in the Fairwind Program. Google describes Carbon as coming soon for general developers while it continues gathering feedback and iterating on guardrails.

Your best operational path is to deploy on Argon today. Design your prompts and API interfaces using standard parameters so you can switch models as soon as Carbon receives a public model identifier.

Gemini 4 Carbon vs Argon Benchmark Gains

In our internal analysis, we compared Google's published scores row by row. Carbon outscores Argon on all 17 rows.

The largest performance separations appear in security tasks, long-context retrieval, and scientific terminal operations. In contrast, Chartography, Vals Index v2.2 and AutomationBench show gains of 2.6 points or less.

Readers can review the complete dataset across all evaluated competitors in our full benchmark table.

Access and Availability Differences

Gemini 4 Argon is active on the Gemini API under the model name gemini-4-argon. Developers can configure calls immediately on the Gemini API as explained in our guide on how to use Gemini 4 Argon.

Gemini 4 Carbon has no active model identifier on the official Gemini API models documentation. Google is rolling out Carbon access exclusively to select cyber defenders and critical infrastructure organizations, with additional details covered in our Gemini 4 Carbon explained guide.

Organizations that qualify as critical infrastructure or defensive security researchers can submit an application using the official Fairwind Program interest form. Artificial Analysis lists Carbon as available through one provider at $4 input and $20 output per 1M tokens, but Google's official documentation continues to designate Carbon access as limited to Fairwind partners.

Token Pricing and Billing Timelines

Both models share identical introductory pricing on standard API calls. Google established an introductory price of $2 per 1M input tokens, $10 per 1M output tokens, and $0.10 per 1M cached input tokens for both Argon and Carbon.

The difference lies in the price transition timeline. Argon's promotional pricing ends on November 30, 2026, jumping to $4 per 1M input, $20 per 1M output, and $0.20 per 1M cached tokens on December 1, 2026, as documented on the Gemini API pricing page. Additional rate details are analyzed in our Gemini 4 Argon pricing breakdown.

Google has not published the end date for Carbon's introductory promotional window. Google's blog announcement puts Carbon's post-introductory rates at $4 input and $20 output per 1M tokens. Google has not confirmed whether the 95% caching discount continues, which would make cached input $0.20. Further pricing implications are examined in our Gemini 4 Carbon pricing guide.

→ Compare per-token pricing for every AI model

What Gemini 4 Argon Offers Today

Gemini 4 Argon has a published spec sheet that developers can build on today. Google lists each of these on the official Argon model page.

Argon supports a unified context window of 1,048,576 tokens covering inputs across text, images, video, audio, and PDF documents. Output length defaults to 65,536 tokens, but can reach up to 1,048,576 tokens when using long decode.

Argon includes thinking controls with low, medium, and high levels, function calling, structured outputs, code execution, file search, URL context, Google Search grounding, Google Maps grounding, and experimental preview access to computer use. Argon also supports Batch API, Flex and Priority inference, with implicit caching only. Google has published none of these specifications for Gemini 4 Carbon.

Ways to Cut Gemini 4 Argon Cost While You Wait

Production teams waiting for Carbon can optimize their spend on Argon using alternative API processing tiers. Argon offers Batch API and Flex inference modes at $1.00 per 1M input tokens and $5.00 per 1M output tokens through November 30, 2026.

After November 30, Batch and Flex rates move to $2.00 input and $10.00 output per 1M tokens. Priority processing currently costs $3.60 input and $18.00 output, before shifting to $7.20 input and $36.00 output on December 1.

Implicit caching bills cached input at $0.10 per 1M tokens through November 30, 2026, and $0.20 from December 1. For example, batch processing 1M input tokens and 200k output tokens costs exactly $2.00 through November 30, calculated as 1 × $1.00 plus 0.2 × $5.00. Google has not published Batch or Flex pricing structures for Gemini 4 Carbon.

→ Compare per-token pricing for every AI model

Which Workloads Gain Most from Gemini 4 Carbon

Defensive cybersecurity and vulnerability research workloads show the largest jump on Carbon. Carbon achieved 54.8% on ExploitGym compared to Argon's 42.0%, representing a gain of 12.8 percentage points when evaluated without guardrails.

Long-context graph traversal also gains a lot on Carbon. On the GraphWalks benchmark evaluating breadth-first search across 256k to 1M tokens, Carbon hit 96.2% F1 versus 84.2% for Argon, establishing a 12.0 percentage point gain.

Terminal execution and agentic environments show comparable progress. Carbon scored 64.8 on Terminal-Bench-Science against Argon's 52.9, while SWE-sweep scores more than doubled from 4.3% on Argon to 9.6% on Carbon. Workloads centered on chart reading or simple task automation show modest gains, such as a 0.9 point rise on Chartography and a 2.6 percentage point increase on AutomationBench.

Safety Differences Google Published

Google published safety results comparing both models directly. On the Gray Swan indirect prompt-injection evaluation, Carbon recorded an indirect prompt-injection attack success rate of 0.0%, compared to 0.7% for Argon.

Google's internal assessments report that Carbon delivers more than a 2x uplift in its success rate for generating secure code compared to Argon. Google also claims Carbon reduces unsupported statements by approximately 2x over Argon.

These internal measurements reflect Google's own testing evaluations rather than independent audits. Carbon's cybersecurity benchmarks like ExploitGym were specifically evaluated without guardrails.

What Google Has Not Published for Gemini 4 Carbon

Google has not published several technical specifications and commercial terms for Gemini 4 Carbon that are already documented for Gemini 4 Argon.

Re-check the Gemini API models page for a Carbon model ID before planning production deployments.

• Context window and output limit: Carbon lists no context window or output token limit, while Argon provides a 1,048,576-token unified context and a 65,536 default output limit.
• Gemini API model ID: Google has not published a model ID for Carbon, whereas Argon uses the model ID gemini-4-argon.
• Rate limits: Google has not published request or token rate limits for Carbon.
• Platform availability: Google has not confirmed whether Carbon will be available in the Gemini app, Google AI Studio, Vertex AI, or any Google AI plan.
• Introductory pricing duration: The end date for Carbon introductory pricing remains unannounced, while Argon introductory rates run through November 30, 2026.
• Post-introductory cached inputs: Google lists no post-introductory cached-input rate for Carbon, whereas Argon charges $0.20 from December 1, 2026.
• Batch, Flex, or Priority rates: Carbon lists no Batch, Flex, or Priority rates, while Argon lists Batch and Flex rates of $1.00 for inputs and $5.00 for outputs through November 30, 2026.
• Free tier: Google has not announced a free tier for Carbon, and Argon offers no free tier.

Who Should Not Wait for Gemini 4 Carbon

Teams with active development deadlines should not wait for Carbon. Google has given no public release date for Gemini 4 Carbon, while Gemini 4 Argon is available today on the Gemini API.

Regulated organizations requiring documented availability should also stick with Argon. Argon provides published technical limits and documented availability that Carbon currently lacks.

Two specific developments would change our guidance. If Google issues a public Gemini API model identifier for Carbon at the launch price, teams handling long-context retrieval and security analysis should switch immediately. Google retiring Gemini 4 Argon would also change this guidance, and Google has announced no such plan.

The Verdict

Choose Gemini 4 Argon for all current production systems. Argon is generally available on the Gemini API, provides documented 1M-token context handling, and shares the same $2 input and $10 output introductory pricing that Carbon will use.

Plan to switch to Gemini 4 Carbon when Google opens public developer access, especially for security work, terminal agents or long-context retrieval, where Carbon gained 11.9 to 12.8 points over Argon.

Build on Gemini 4 Argon today, keep prompts free of model-specific wording, and if you run your own evaluations, rerun them on Gemini 4 Carbon once public API access opens.