Gemini 4 Carbon goes GA at a quarter of Fable's price, with a swarm attached
$2 per million input tokens, $10 per million output. One day after Anthropic priced Claude Fable 5.2 at $8/$40, Google opened Gemini 4 Carbon to every paying developer and enterprise at the same introductory price as Argon. The footnote matters: once the introductory period ends on November 30, 2026, the price becomes $4/$20. So the gap with Fable halves in three weeks, but it is still a 2x gap.
Two weeks ago Carbon went only to cyber defenders through the Fairwind Program, which is what we covered then. Now it is in AI Studio, Gemini for business, and rolling out to AI Ultra subscribers in Antigravity, the Gemini app and AI Mode. Google is unusually clear about positioning: Argon stays the default for most coding and knowledge work, and Carbon is the one that burns more tokens on "extended verification loops" for hard multi-step jobs.
The more interesting launch is the preview of /swarm in Antigravity. One prompt becomes a coordinated team of agents you can watch in real time. Carbon takes the roles that need the deepest reasoning, architecture, coordination and auditing, while faster, lighter models work in parallel isolated workspaces and adversarial verifiers test every change. Read that next to Fable 5.2's launch post, which pitched Fable as an Advisor or Orchestrator over Haiku, Sonnet and Opus. Two labs, 24 hours apart, same pitch: the top model is the manager, not the worker. Pricing per token now matters less than pricing per managed team.
Google's benchmark claims are all cost-framed. Carbon is #1 on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench "at more than 2X lower cost per task" than the next best external model, and #2 on Artificial Analysis's new Multimodal Knowledge Agent benchmark at more than 6x lower average cost per task than the leader. AutomationBench is the one with an outside cross-check: Anthropic's own Fable 5.2 table already conceded Carbon's 53.9% against Fable's 51.4%. The rest is Google's comparison set, with Google choosing which "external model" counts as next best.
On safety, Google says Carbon blocked 99% of attack trajectories from its adaptive red-teaming agent and passed 92% of a 375-task agentic honesty benchmark, against 70% for Argon. The honest read: the price war at the top is now a war over who runs your agent team most cheaply, and Google just undercut on the manager's salary.
Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/
← Back to all articles
Two weeks ago Carbon went only to cyber defenders through the Fairwind Program, which is what we covered then. Now it is in AI Studio, Gemini for business, and rolling out to AI Ultra subscribers in Antigravity, the Gemini app and AI Mode. Google is unusually clear about positioning: Argon stays the default for most coding and knowledge work, and Carbon is the one that burns more tokens on "extended verification loops" for hard multi-step jobs.
The more interesting launch is the preview of /swarm in Antigravity. One prompt becomes a coordinated team of agents you can watch in real time. Carbon takes the roles that need the deepest reasoning, architecture, coordination and auditing, while faster, lighter models work in parallel isolated workspaces and adversarial verifiers test every change. Read that next to Fable 5.2's launch post, which pitched Fable as an Advisor or Orchestrator over Haiku, Sonnet and Opus. Two labs, 24 hours apart, same pitch: the top model is the manager, not the worker. Pricing per token now matters less than pricing per managed team.
Google's benchmark claims are all cost-framed. Carbon is #1 on CursorBench 4.0, Perplexity's WANDR and Zapier's AutomationBench "at more than 2X lower cost per task" than the next best external model, and #2 on Artificial Analysis's new Multimodal Knowledge Agent benchmark at more than 6x lower average cost per task than the leader. AutomationBench is the one with an outside cross-check: Anthropic's own Fable 5.2 table already conceded Carbon's 53.9% against Fable's 51.4%. The rest is Google's comparison set, with Google choosing which "external model" counts as next best.
On safety, Google says Carbon blocked 99% of attack trajectories from its adaptive red-teaming agent and passed 92% of a 375-task agentic honesty benchmark, against 70% for Argon. The honest read: the price war at the top is now a war over who runs your agent team most cheaply, and Google just undercut on the manager's salary.
Source: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/
Comments