AI model prices, context windows & deprecations.

Google's Gemini API release notes for November 12, 2026 read 'Gemini 4 Hydrogen (gemini-4-hydrogen) generally available (GA)', and the model page lists 'Stable: gemini-4-hydrogen'. Available on the free tier and at paid-tier Standard $1.50 input / $7.50 output per 1M tokens (output including thinking tokens), implicit caching $0.075; Batch and Flex $0.75 / $3.75, Priority $2.70 / $13.50. No dated price schedule is published for it, unlike Gemini 4 Carbon and Argon. 1,048,576-token unified context window, 65,536 max output by default and up to 1,048,576 with long decode; text, image, video, audio and PDF input, text output; adds an xhigh thinking level above the default high. Google calls it the recommended upgrade from Gemini 3.8 Flash and earlier Gemini 3 and 2.5 models and makes it the default for the Antigravity agent, but declares no shutdown for Gemini 3.8 Flash. No knowledge cutoff is published. Recorded one day after Google's date.

Google's Gemini API release notes for November 11, 2026 announce Gemini Decision (gemini-decision-1) and the Decisions API 'for fast, calibrated binary, categorical, and rubric decisions across text, images, audio, video, and PDFs with zero output tokens'. A single request can evaluate up to 1,000 questions over up to 262,144 tokens of shared input and returns probabilities rather than generated text. Paid-tier pricing is $0.10 per 1M input tokens for every input modality; output is 'Not applicable (0 output tokens)', context caching is not available and the Batch API is not supported. The deprecations page lists it with 'No shutdown date announced' and no replacement. Google does not label it Stable or Preview; it is catalogued as ga on Google's naming convention (no -preview suffix), which the model row explains. Recorded one day after Google's date.

Google's Gemini API release notes for November 10, 2026 read 'Gemini 4 Carbon (gemini-4-carbon) generally available (GA)', calling it 'our most capable reasoning model built for sustained agentic coding, multimodal knowledge work, scientific reasoning, and cybersecurity defense', and the model page lists 'Stable: gemini-4-carbon'. Paid tier only. The pricing page publishes a dated two-column schedule: Standard $2.00 input / $10.00 output per 1M tokens (output including thinking tokens) with implicit caching at $0.10 through November 30, 2026, then $4.00 / $20.00 / $0.20 starting December 1, 2026 - the same schedule, cell for cell, as gemini-4-argon, which Google now describes as 'our efficient Gemini 4 frontier reasoning model' with 'lower token spend per turn compared to Gemini 4 Carbon'. Batch and Flex are half the Standard rate, Priority 1.8x. 1,048,576-token unified context window, 65,536 max output by default and up to 1,048,576 with long decode; text, image, video, audio and PDF input, text output. No knowledge cutoff is published, and Google names no model that Carbon replaces. Recorded one day after Google's date.

OpenAI's deprecations page added the section "2026-11-10: GPT Image 2": "The gpt-image-2 model and gpt-image-2-2026-04-21 snapshot are deprecated and will be removed from the API on February 10, 2027." The recommended replacement is given as "gpt-image-2.6-sunburst or gpt-image-2.6-flare", and the gpt-image-2 model page now reads "Status: Deprecated". Three months is the minimum notice OpenAI's own policy states for image generation models. In the same pass OpenAI re-pointed the older '2026-06-02: GPT Image model deprecations' table: gpt-image-1-mini, gpt-image-1.5 and the chatgpt-image-latest alias (all shutting down December 1, 2026) now list the two GPT Image 2.6 models as their replacement where they previously listed gpt-image-2. This catalog's replacement field holds one id and carries gpt-image-2-6-sunburst, the first-named; Flare is priced identically. WHAT THIS IS NOT: not a shutdown - gpt-image-2 is still priced at $8 image input / $30 image output per 1M tokens and the removal date is three months out. GPT Image 2.5 Sunburst and Flare are not named in either table and are not deprecated.