# Google Releases Gemini 4 Neon at $3 per Million Input Tokens, a Fraction of Fable 5.2's Price

Google says Neon ranks third on the Artificial Analysis Intelligence Index with a score of 61, but that score uses an "xhigh" reasoning setting that users of the Gemini app cannot select.

- Published: 2026-11-19T06:16:23.269Z
- Canonical: https://polylog.news/ai/2026-11-19/google-releases-gemini-4-neon-at-3-per-million-input-tokens
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Google (via Hacker News)](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/), [Polylog editors](https://polylog.news), [Artificial Analysis](https://artificialanalysis.ai/models/gemini-4-neon), [Plotly](https://plotly.com/blog/gemini-4-neon-plotly-data-analytics-bench/), [Digital Trends](https://www.digitaltrends.com/computing/google-says-gemini-4-neon-can-hang-with-claude-opus-5-5-and-gpt-6-2-sol-for-less/), [AI Release Tracker](https://aireleasetracker.com/model/google/gemini-4-neon)

Google released [Gemini 4 Neon](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/) on November 18. In the Gemini 4 family it ranks between Carbon, which Google positions for complex, long-horizon projects, and Hydrogen, which it positions for fast everyday tasks. It has a 1 million token context window and [four reasoning-depth levels](https://t.me/ai_machinelearning_big_data/11385). Pricing is $3 per million input tokens and $15 per million output tokens.

The headline claim depends on one benchmark. [Artificial Analysis](https://artificialanalysis.ai/models/gemini-4-neon) scores Neon at 61 on its Intelligence Index at the xhigh setting, 60 at high, 55 at medium and 44 at low. On that basis Google places Neon third, behind Claude Fable 5.2 at 65 and GPT-6.2 Astra at 62. Google's comparison table puts Neon ahead of Claude Opus 5.5 (59) and GPT-6.2 Sol (57). Google acknowledges that Neon trails Opus 5.5 on Terminal-Bench 4.1 and FrontierCode 2.0 and trails GPT-6.2 Sol on Chartography, [according to Digital Trends](https://www.digitaltrends.com/computing/google-says-gemini-4-neon-can-hang-with-claude-opus-5-5-and-gpt-6-2-sol-for-less/).

Google reports 49.1% on SWE-bench Pro, 62.4% on Terminal-Bench 2.0 and 72.3% on BrowseComp. Those figures come from Google's own harness. An [independent tracker](https://aireleasetracker.com/model/google/gemini-4-neon) lists 46.2% on FrontierCode 2.0 at the default high setting and notes that Google's harness produced a different score. That puts Neon about three points below Fable 5.2's vendor-reported 49.1% on the same test.

[Plotly's data-analytics benchmark](https://plotly.com/blog/gemini-4-neon-plotly-data-analytics-bench/) supplies a third-party result. Neon passed 42 of 43 questions for $17.03, matching Gemini 4 Carbon. Its median question took 46 seconds, against 18 seconds for GPT-6.1 Sol and 93 seconds for Carbon. On that test the model costs less but runs slower.

## What this means

The mechanism is price per unit of near-frontier capability. Neon's list price is 37.5% of Fable 5.2's on both input and output. If Neon's score at xhigh holds up independently, it is within four index points of the leader. The exposed parties are Anthropic's API gross margin, on the day its stock lists, and OpenAI's Sol tier. Meta's Muse Spark 1.1 is already priced lower, at $1.25 and $4.25, but sits at a lower capability tier. The market expected Google to compete on price, given its own TPUs. What is new is a model that nearly matches the leader's score while costing far less. The consensus treats index rank as the decision variable. For agentic coding buyers, the numbers that matter are FrontierCode and Terminal-Bench, where Neon still trails Opus 5.5. It also matters that the index's top setting raises token use and latency. The desk's read is that Anthropic or OpenAI responds within weeks with a cheaper mid-tier model or a repricing, provided independent harnesses confirm Neon within a few points of Fable 5.2 on coding. A larger independent gap on agentic coding tasks would keep the premium tier's pricing intact and prove this read wrong.

## What to watch

- Independent Terminal-Bench and FrontierCode runs of Neon at the high setting, because a gap of more than a few points from Fable 5.2 would weaken Google's price argument for coding buyers.
- Any price cut or new mid-tier model from Anthropic or OpenAI in the coming weeks, which would confirm that Neon has reset the reference price.
- Per-task token use at the xhigh setting, because a cheap per-token price can still produce expensive tasks if the model reasons for longer.
