Google's Gemini 4 Neon nearly ties Fable 5.2 on Vals Index
Google's new model lands within the margin of error of Claude Fable 5.2, though Artificial Analysis clocks it as slow at its highest setting.

In short
- 01Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window.
- 02Gemini 4 Carbon's API price doubles to $4 and $20 starting Dec. 1, which makes Neon 25% cheaper per token.
- 03Artificial Analysis measured Gemini 4 Neon at 22.8 minutes per task at xhigh, against a 12.7-minute median for comparable models.
The story
Google just launched Gemini 4 Neon, a model that sits between its quick Hydrogen and its strongest model, Gemini 4 Carbon. Neon costs $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window.
Independent scores back up much of the pitch. On the Vals Index, Neon scores 70.87%, just behind Anthropic's Claude Fable 5.2 at 71.31%. Each score carries a margin of about one point, so the gap sits inside the margin of error.
Neon becomes cheaper than Carbon starting Dec. 1
Carbon still runs at an introductory $2 for input and $10 for output. Google's pricing page shows those rates rising to $4 and $20 starting Dec. 1, which puts Neon 25% below Carbon per token.
VentureBeat's price comparison puts Neon's rates 50% above OpenAI's GPT-6.2 Sol and Claude Sonnet 5.5, which both charge $2 and $10. Neon also comes in 25% below Claude Opus 5.5, which charges $4 and $20.
Artificial Analysis finds Neon slow at its highest setting
Google promotes fast task completion, but Artificial Analysis measured 22.8 minutes per task at the top xhigh setting, against a 12.7-minute median for comparable models. At the default high setting, Neon took 14.8 minutes.
The xhigh run also produced 283 million output tokens, against a median of 65.4 million, and cost $6.44 per task on the firm's index.
Google also reports its lowest fabrication rate yet, but that figure comes from its own internal testing rather than an outside benchmark.
Developers are struggling to place Neon in Google's lineup
The Hacker News launch thread drew 775 comments. Of the top 30 when we read them, about a third tried to work out where Neon fits or complained about Google's model names. That's a snapshot, not a survey.
I'm getting fatigued with the constant introduction of names and the impenetrability of mapping these brand new names to use cases.
stillpointlab on Hacker News
Neon is rolling out on the paid Gemini API through Google AI Studio and in Gemini for business. Google AI Pro and Ultra subscribers get it in the Gemini app and Google Antigravity. It isn't available on the API's free tier.
Why it matters
Google now has a mid-priced model near the top of an independent index, and the Carbon price rise gives cost-conscious teams a reason to move. The catch is time and tokens. At its highest setting, Neon works slowly and writes a lot, so the per-token price alone understates what a long agent run will cost.
Tools in this story
Share


