News

Gemini 4 Neon Released: $3 / $15 per Million Tokens, and How It Compares With Claude Opus 5.5 and GPT-6.2 Sol

Google released Gemini 4 Neon on November 18, 2026. It is a new middle tier in the Gemini 4 family: Google’s model card says it is adapted from Gemini 4 Carbon and sits between Carbon, “for your hardest, long-horizon projects”, and Hydrogen, for “fast everyday execution”. Through the API it costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window.

According to Google’s launch post, Neon is rolling out in the Gemini API and Google AI Studio, in Gemini for business, in Google Antigravity for paid Google AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. It is not available on the free tier of the API.

Prices per 1M tokens

From the Gemini API pricing page, checked November 19, 2026:

Tier Input Cached input Output (including thinking)
Standard $3.00 $0.15 $15.00
Batch $1.50 $0.075 $7.50
Flex $1.50 $0.075 $7.50
Priority $5.40 $0.27 $27.00

Caching is implicit only. A request with 1,000 input and 500 output tokens costs about $0.0105 at standard rates. Thinking tokens are billed as output, so a hard task at a high thinking level costs more than the visible answer suggests. Grounding with Google Search is free for 5,000 requests a month across Gemini 3 and newer models, then $14 per 1,000.

How the price compares

Model Input Output Note
Gemini 4 Hydrogen $1.50 $7.50
GPT-6.2 Sol $2.00 $10.00
Gemini 4 Carbon $2.00 $10.00 Launch price through November 30, then $4 / $20
Gemini 4 Argon $2.00 $10.00 Launch price through November 30, then $4 / $20
Gemini 4 Neon $3.00 $15.00
Claude Opus 5.5 $4.00 $20.00

Sources: Gemini API pricing, OpenAI pricing, Anthropic pricing, all checked November 19, 2026.

Two things stand out. Until November 30, Neon costs more per token than Gemini 4 Carbon, because Carbon is still on its launch price. From December 1, when Carbon goes to $4 / $20, Neon is 25% cheaper per token. And GPT-6.2 Sol is a third cheaper than Neon per token, while Claude Opus 5.5 costs a third more.

Google’s own headline figure is different: it says that at the default “high” thinking level Neon costs 32% less per task than Carbon. That is a cost per completed task, which also depends on how many tokens each model uses, not a price per token. It is Google’s measurement.

Benchmarks: what Google’s table shows

Google compared Neon with Claude Opus 5.5 and GPT-6.2 Sol in a results table on 19 benchmark rows. Neon has the best score in 16 of them. Opus 5.5 is ahead on two coding tests and GPT-6.2 Sol on one chart-reading test:

Benchmark Gemini 4 Neon Claude Opus 5.5 GPT-6.2 Sol
Artificial Analysis Intelligence Index v5.0.1 61 59 57
Vals Index v2.2 70.9% 70.1% 66.1%
Terminal-Bench 4.1 (agentic coding) 65.2% 65.8% 61.5%
FrontierCode 2.0 (coding) 46.2% 47.3% 43.8%
CursorBench 4.1 (coding) 65.4% 58.1% not reported
AutomationBench (knowledge work) 56.3% 42.5% 41.3%
HLE-Diamond (reasoning) 71.3% 67.9% 64.0%
Chartography (reading charts) 73.5% 66.3% 75.1%
GraphWalks, 256K to 1M tokens (long context) 95.2% 66.8% 80.1%

The best score in each row is in bold. Selected rows; the full table has 19.

Read these numbers with three caveats, all from Google’s own methodology note:

  • The rivals’ numbers are mostly self-reported by Anthropic and OpenAI or taken from public leaderboards, while several Neon scores were computed by Google, some in its own Antigravity harness. On SWE-sweep, Neon scored 39.0% in Antigravity against 10.1% for Opus 5.5 in Claude Code, but only 12.8% in a neutral harness.
  • The comparison leaves out bigger models. On the Artificial Analysis Intelligence Index, Google says Neon ranks third overall, “just behind larger, slower models”. It is the best of the three models in the table, not the best model available.
  • Settings differ. Google ran Neon at its highest thinking level for most tests but at “high” for FrontierCode 2.0, and used the best available reasoning settings for the rivals.

Google also says Neon ranks first on the Artificial Analysis Coding Agent Index (58.6) and, in some security tests, beats Gemini 4 Carbon: 70.7% on CWE-bench v2, 3.4 points above Carbon, and more secure code than Carbon on AutoBaxBench. These are vendor claims we have not checked independently.

Other details

  • Context and output. 1,048,576 tokens of context shared between input and output. Output is 64K tokens by default and up to 1 million with Google’s long decode mode.
  • Knowledge cutoff. March 2026, according to the model card.
  • Thinking levels. Four levels; “high” is the default and “xhigh” the deepest.
  • Cyber version. A separate Gemini 4 Neon for cyber is going to “trusted defenders” through Google’s Fairwind Program “in the coming days”.
  • Safety report. The full Safety and Security Report is “coming soon”. The model card says Neon did not cross a new critical capability threshold compared with Gemini 4 Argon, so Google carried forward Argon’s risk assessments.

Should you switch?

If you use Gemini 4 Carbon or Argon through the API, Neon is not cheaper per token until December 1, so there is no rush for cost reasons. After that, it is worth testing on your own workload, since Google’s claim is lower cost per task at similar quality. If you choose between vendors, GPT-6.2 Sol is cheaper per token and Claude Opus 5.5 is more expensive, and Google’s table puts Neon ahead of both on most of its chosen tests. Test on your own prompts before you switch: vendor benchmarks are chosen by the vendor.

Current per-model prices for every vendor are in our LLM API pricing table, and the model page for Gemini 4 Neon lists the sources behind each number. For a side-by-side view, see Gemini 4 Neon vs Claude Opus 5.5, Gemini 4 Neon vs GPT-6.2 Sol and Gemini 4 Neon vs Gemini 4 Argon. Price changes are tracked in the model changelog.