Gemini 4 Carbon

GA/stable ('Stable: gemini-4-carbon' on the model page; the November 10, 2026 release note is headed 'Gemini 4 Carbon (gemini-4-carbon) generally available (GA)'), described by Google as 'our most capable reasoning model, built for complex software engineering, mathematical and scientific problem solving, high factual accuracy, autonomous multi-step agents, and long decode up to 1M tokens'. THE PRICE FIELDS CARRY THE INTRODUCTORY COLUMN IN FORCE THROUGH 2026-11-30, not the later one - Google's pricing page publishes a two-date schedule: paid-tier Standard is $2.00 in / $10.00 out per 1M (output including thinking tokens) with context caching $0.10/1M through November 30, 2026, doubling to $4.00 / $20.00 / $0.20 starting December 1, 2026; the release note and the 'What's new in Gemini 4 Carbon' page both call the first column introductory pricing. A midnight boundary, so there is no intra-day error window, but the fields must be re-read on 2026-12-01: since Google moved its docs to aistudio.google.com on 2026-11-03, scripts/check-google-prices.mjs parses no rate on that page and will NOT flag this cutover on its own. Other published tiers, none of which fit the per-MTok fields, on the same two dates: Batch and Flex $1.00 in / $5.00 out (caching $0.05) then $2.00 / $10.00 (caching $0.10); Priority $3.60 / $18.00 (caching $0.18) then $7.20 / $36.00 (caching $0.36). The schedule is identical, cell for cell, to gemini-4-argon's; the What's-new page says Carbon is offered 'at the same per-token price' as Argon. Caching is implicit only - Google states explicit caching is not supported, so no cache storage fee is published. Grounding with Google Search or Maps: 5,000 requests/month free shared across all Gemini 3 and newer models, then $14 per 1,000. Not available on the Free tier (every free-tier cell reads 'Not available'). Spec table gives a 1,048,576-token context window that Google labels a 'unified input and output token limit', and max output 65,536 (default synchronous) / up to 1,048,576 with long decode - max_output_tokens carries the default synchronous figure. Inputs text/image/video/audio/PDF, output text; thinking supported at low, medium and high (default medium per the What's-new page); computer use, code execution, file search, function calling and structured outputs supported; Live API, audio generation and image generation not supported. Released 2026-11-10 per the API release notes and the 'Gemini 4 models' table of the deprecations page, which lists it with 'No shutdown date announced' and no replacement; the model page's 'Latest update' field carries the date more coarsely as 'November 2026'. Google publishes no knowledge cutoff on the model page, pricing page, What's-new page or release note, so that stays null. Google names no model that Carbon replaces. First read 2026-11-11.

Source: Google official documentation ↗

Explore more

Gemini 4 Carbon — questions & answers