Carbon, Argon and Sonnet 5.5: what the model listings and API prices show

Google’s Gemini 4 Carbon, announced October 28, is at a different stage from two models already listed in provider API documentation. Google describes initial access for selected cyber defenders, followed by trusted Fairwind partners; wider developer, business and consumer access is still forthcoming. It has not established a public API model ID or current public API price in the material reviewed.

For Gemini 4 Argon, Google announced developer and enterprise API availability on October 13. Its documentation lists the model as stable under the API code `gemini-4-argon`. Anthropic lists Claude Sonnet 5.5 as active, released September 28, and available through the Claude API and cloud platforms. Its API ID is `claude-sonnet-5-5`. These are provider descriptions: an individual account’s access has not been tested.

The posted text-token rates observed on October 31 are in US dollars and give developers a starting point for comparing API charges. Google’s paid-tier Argon Standard rate is $2 per million input tokens and $10 per million output tokens through November 30, 2026; output includes thinking. Its asynchronous Batch rate is $1 input and $5 output per million tokens. Google says that from December 1 the Standard rates will rise to $4 and $20, and Batch to $2 and $10, respectively. The effective time zone was not specified.

Anthropic’s first-party Claude API lists Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens for Standard, and $1 and $5 for asynchronous Batch. The effective date of those prices was not established. Where supported, US-only inference routing multiplies token rates by 1.1.

Rowan Finch analysis

An announcement can describe a rollout without making a model available to every account. A documented API release is more specific, but access still needs to be checked for the account and service in question. Model IDs matter too: Anthropic’s versioning documentation describes the dateless Sonnet ID as a pinned snapshot, while Google’s named Argon release is not established here as an immutable snapshot.

The observed base text-token rates match today, but that does not make the cost of completing a particular job equal or establish equal performance. Workloads can use different numbers of tokens, and Google’s announced December increase makes the date part of the comparison.

Source note: This overview uses Google and Anthropic’s own announcements, model documentation and price pages. The rates exclude extras such as tools, taxes and negotiated or partner terms; neither access, billing nor performance was independently tested.