What will Gemini 4 Carbon cost after 1 December? Neon, GPT-6.2 Sol and Claude compared
Gemini 4 Carbon and Argon double in price on 1 December 2026. What one monthly workload costs on Carbon, Neon, GPT-6.2 Sol, Claude and Grok at list prices.
By Emre A.Published
Tech AI LatestOn 1 December 2026 Google moves Gemini 4 Carbon and Argon from $2 and $10 to $4 and $20 per million tokens. At list prices, a workload of 10 million input and 2 million output tokens a month rises from $40 to $80. The same workload costs $60 on Gemini 4 Neon, $40 on GPT-6.2 Sol or Claude Sonnet 5.5, $30 on Gemini 4 Hydrogen and $160 on GPT-6.2 Astra or Claude Fable 5.2.
Key facts
- Price change: Gemini 4 Carbon and Argon go from $2 and $10 to $4 and $20 per 1M input and output tokens on 1 December 2026 (Gemini API pricing)
- Google's new middle option: Gemini 4 Neon, released 18 November at $3 and $15, adapted from Carbon
- Same price as Carbon today: GPT-6.2 Sol and Claude Sonnet 5.5, both $2 and $10
- Same price as Carbon from December: Claude Opus 5.5 at $4 and $20
- Flagships: GPT-6.2 Astra and Claude Fable 5.2 both list at $8 and $40
- All figures are list prices per 1M tokens; token use per task differs by model and thinking level
What changes on 1 December
Google launched Gemini 4 Argon on 13 October and opened Gemini 4 Carbon on the paid Gemini API on 10 November, both at an introductory $2 per million input tokens and $10 per million output tokens. The Gemini API pricing page and Google's Carbon announcement set the end of that offer: from 1 December 2026, both models cost $4 and $20. For any team that built on Carbon or Argon in the past six weeks, the per-token bill doubles in eleven days.
Google did not leave those customers without an exit. On 18 November it released Gemini 4 Neon, which its model card describes as adapted from Gemini 4 Carbon, at $3 and $15 per million tokens. A week earlier it had launched Gemini 4 Hydrogen at $1.50 and $7.50. The Gemini 4 line therefore now has three price points after December: Hydrogen, Neon, and Carbon and Argon at the top.
The question for developers is not only which Gemini model to use, but whether the price change is the moment to compare the whole market again. In the same fortnight, OpenAI released GPT-6.2 Astra and Sol, and Anthropic cut the price of its flagship with Claude Fable 5.2. This piece puts the published list prices side by side and runs one example workload through them.
List prices side by side
All prices below are per million tokens, input then output, as each developer published them. Gemini 4 Hydrogen costs $1.50 and $7.50, flat across its 1,048,576 token context window, according to Google's Hydrogen announcement. Grok 4.7 costs $2 and $6 below 200,000 prompt tokens and $4 and $12 above, according to the xAI release notes. GPT-6.2 Sol costs $2 and $10, unchanged from GPT-6.1 Sol, according to OpenAI. Claude Sonnet 5.5 also costs $2 and $10, according to Anthropic.
In the middle, Gemini 4 Neon costs $3 and $15 per million tokens, according to 9to5Google and other outlets reporting the figures from Google's launch. Claude Opus 5.5 costs $4 and $20, according to Anthropic, which from 1 December is exactly the price of Gemini 4 Carbon and Gemini 4 Argon. At the top, GPT-6.2 Astra and Claude Fable 5.2 both list at $8 and $40, after OpenAI cut Astra by 20% on 17 November and Anthropic cut Fable by 20% on 9 November.
Cached input changes the picture for workloads that reuse long prompts. Gemini 4 Neon charges $0.15 per million cached input tokens, GPT-6.2 Sol $0.10, Claude Sonnet 5.5 $0.10 for cache reads, Claude Opus 5.5 and Claude Fable 5.2 $0.20, GPT-6.2 Astra $0.40 and Grok 4.7 $0.50 below its 200,000 token threshold. These rates come from the same announcements and pricing pages. Our model release tracker lists every row with its source.
What one workload costs on each model
To make the prices comparable, take a modest production workload: 10 million input tokens and 2 million output tokens a month, with no caching, no batch discount and prompts below 200,000 tokens. The arithmetic below is ours, at list prices.
On Gemini 4 Carbon or Argon, that workload costs $40 a month in November ($20 for input, $20 for output) and $80 from 1 December. On Gemini 4 Neon it costs $60. On Gemini 4 Hydrogen it costs $30. On GPT-6.2 Sol and on Claude Sonnet 5.5 it costs $40, the same as Carbon today. On Grok 4.7 it costs $32, because of its lower output price. On Claude Opus 5.5 it costs $80, the same as Carbon after the change. On GPT-6.2 Astra and Claude Fable 5.2 it costs $160.
Three things follow. First, a Carbon customer who does nothing pays twice as much from December for the same tokens. Second, moving to Neon cuts that new bill by 25%, but still costs 50% more than Carbon did in November. Third, moving outside Google to GPT-6.2 Sol or Claude Sonnet 5.5 keeps the November bill unchanged at list price, if the quality is good enough for the task. A workload heavy on output tokens would widen the gaps, because output prices differ more than input prices.
Per-token prices are only half of the bill. Gemini 4 models offer several thinking levels, and Claude models several effort settings; higher levels spend more output tokens on each task. OpenAI says GPT-6.2 Astra used 62% fewer tokens than GPT-6.1 Sol on complex legal agent tasks in its own tests. Vendor token claims are measured on tasks the vendor chose, so the only reliable number is a test run on your own prompts.
What you get for the price
Benchmarks published by the vendors are not neutral, but Google's Neon launch is useful because it put Neon next to two rivals on the same tests. On Google's figures, Neon scores 70.5% on Terminal-Bench Science against 68.0% for GPT-6.2 Sol and 63.3% for Claude Opus 5.5, and 71.3% on HLE-Diamond against 67.9% for Opus 5.5 and 64.0% for Sol. On Terminal-Bench 4.1, the agentic coding test, Opus 5.5 is slightly ahead at 65.8% to Neon's 65.2%, with Sol at 61.5%. These are Google's numbers.
An independent ranking points the same way at the top. Artificial Analysis puts Claude Fable 5.2 first on its Intelligence Index at 65 and GPT-6.2 Astra second at 62, with Sol at 57. The Next Web reported that Artificial Analysis lists Neon at 61 at its highest thinking setting, third on the board. Google reported that Hydrogen scores 51 on the same index at its default high setting and 54 at xhigh. On these figures, Neon sits close to the $8 flagships on capability at well under half their price, while Hydrogen and Sol are the low-cost options with a visible capability gap.
Read together, the prices and scores suggest a rough split. For high-volume, simpler work such as extraction, routing and summaries, Hydrogen, Grok 4.7, Sol and Sonnet 5.5 are the candidates. For complex coding and analysis where Carbon was chosen for quality, Neon and Opus 5.5 are the closest substitutes at or below Carbon's December price. For the hardest long-horizon work, Carbon, Astra and Fable remain the top tier, and the choice there is about task success rather than price per token.
What to check before 1 December
Teams on Carbon or Argon have eleven days to decide. Rerun a sample of real production prompts on Neon and on one model from another vendor, and compare cost per completed task rather than cost per token. Check whether a lower thinking level on Carbon gives acceptable results, because fewer output tokens offset part of the price rise. Check caching: Neon's $0.15 cached input rate matters for agents that resend long system prompts.
Several facts are still missing. Google has not published Neon's separate safety and security report, which the model card lists as coming soon. Google has not said whether Carbon and Argon will get batch or volume discounts after the change. And independent benchmark runs on Neon beyond the Artificial Analysis index are still arriving. The AI model API pricing guide keeps the current table of list prices, all guides are on the guides page, and every release is in AI Models.
Questions people ask
- How much will Gemini 4 Carbon cost from 1 December 2026?
- $4 per million input tokens and $20 per million output tokens, up from the introductory $2 and $10, according to Google's Gemini API pricing page. Gemini 4 Argon changes on the same date to the same prices.
- Is Gemini 4 Neon cheaper than Gemini 4 Carbon?
- From 1 December, yes: Neon costs $3 and $15 per million tokens against Carbon's $4 and $20, a 25% saving. Until 30 November, Neon is 50% more expensive than Carbon's introductory price.
- Which models cost the same as Gemini 4 Carbon does today?
- GPT-6.2 Sol and Claude Sonnet 5.5 both list at $2 per million input tokens and $10 per million output tokens, the same as Carbon's introductory price.
- What does a typical workload cost on these models?
- At list prices, 10 million input and 2 million output tokens a month cost $30 on Gemini 4 Hydrogen, $40 on GPT-6.2 Sol or Sonnet 5.5, $60 on Neon, $80 on Carbon from December or Opus 5.5, and $160 on GPT-6.2 Astra or Claude Fable 5.2.
Sources
- Gemini Developer API pricing (Google, 13 Oct 2026)
- Gemini 4 Carbon is now available (Google, 10 Nov 2026)
- Introducing Gemini 4 Neon (Google, 18 Nov 2026)
- Gemini 4 Neon model card (Google DeepMind, 18 Nov 2026)
- Introducing Gemini 4 Hydrogen (Google, 12 Nov 2026)
- Introducing GPT-6.2 Astra and Sol (OpenAI, 17 Nov 2026)
- Introducing Claude Fable 5.2 and Claude Mythos 5.2 (Anthropic, 9 Nov 2026)
- Introducing Claude Opus 5.5 (Anthropic, 22 Sep 2026)
- Introducing Claude Sonnet 5.5 (Anthropic, 28 Sep 2026)
- xAI API release notes: Grok 4.7, 21 September 2026 (xAI, 21 Sep 2026)
- OpenAI has the #2 model in the Intelligence Index with the release of GPT-6.2 Astra and GPT-6.2 Sol (Artificial Analysis, 17 Nov 2026)
- Google launches Gemini 4 Neon with Carbon quality at Flash speeds, live for AI Pro (9to5Google, 18 Nov 2026)
- Google's Gemini 4 Neon ranks third in AI tests, replacing Argon in Gemini (The Next Web, 18 Nov 2026)
Reference pages
- Gemini 4 Carbon model
- Gemini 4 Argon model
- Gemini 4 Neon model
- Gemini 4 Hydrogen model
- GPT-6.2 Sol model
- GPT-6.2 Astra model
- Claude Sonnet 5.5 model
- Claude Opus 5.5 model
- Claude Fable 5.2 model
- Grok 4.7 model
- Google company
- OpenAI company
- Anthropic company
- xAI (SpaceXSI) company
Spotted an error? See the corrections policy and tell the editor. Facts in this story come from the sources listed above; vendor claims are attributed.
Comments
Loading comments.





