Google's Gemini 4 Neon undercuts Opus 5.5 by 25% and outruns Carbon
Neon lands at $3 and $15 per million tokens, tops Sonnet 5.5 and GPT-6.2 Sol on Google's cited index, and inherits a price edge on Dec. 1.

Google launched Gemini 4 Neon today at $3 per million input and $15 per million output tokens, undercutting Anthropic's flagship Opus 5.5 by 25% while scoring within about a percentage point of it on two engineering benchmarks. For buyers, the launch resets mid-tier model economics just as Carbon's introductory discount expires on Nov. 30, making cost-per-completed-task rather than headline token price the number to watch.
Google launched Gemini 4 Neon today, and the number that matters is 25. Neon's standard rates are $3 per million input tokens and $15 per million output tokens, which Google says sits 25% below Anthropic's flagship Opus 5.5 at $4 and $20, even though Neon scores within about a percentage point of that larger model on two engineering benchmarks. The model is live on the paid Gemini API, in Google AI Studio and Android Studio, and inside the Gemini app and Google's AI-assisted development platform Antigravity for Google AI Pro and Ultra subscribers. Google positions Neon between Hydrogen, its small, light, fast workhorse, and Carbon, the model it recommends for its hardest extended projects.
On the Artificial Analysis Intelligence Index, Neon posts 61 at its xhigh reasoning setting. That is ahead of Opus 5.5 at 59, GPT-6.2 Sol at 57, Claude Sonnet 5.5 at 55, and Google's own Hydrogen at 54, and it edges past Carbon at 60. Artificial Analysis confirmed on X that Neon placed third on its intelligence index. On the Vals Index, which combines finance, software, legal and tax tasks weighted by each sector's contribution to the U.S. economy, Google puts Neon second overall at 70.9% and first on both legal research and tax agent tasks. Two caveats travel with those numbers: Google supplies Neon's Vals result, and the public Vals snapshot predates Neon's release.
The pricing story is stranger than the headline rate suggests. Neon's $3 and $15 are 50% higher than Anthropic's mid-range Claude Sonnet 5.5 and OpenAI's mid-range GPT-6.2 Sol, both at $2 and $10, yet Google reports Neon beats both on the intelligence index at its highest reasoning setting. Against Carbon, the calendar does the work. Carbon's introductory rate of $2 and $10 makes it cheaper per token than Neon through Nov. 30. Beginning Dec. 1, Carbon moves to $4 and $20, and Neon's input and output rates land 25% below Carbon's scheduled rates. Long context scrambles the OpenAI comparison too: under OpenAI's long-context pricing, a Sol request exceeding 272,000 input tokens costs $4 per million input and $15 per million output for the whole request, which erases Sol's headline output-price advantage at that length.
Speed is where Neon separates from its own family. Once it starts generating output, Neon streams about 110 tokens per second at xhigh, according to the launch-day reading from benchmark tracker Priors. Artificial Analysis lists Carbon at 59 tokens per second on high, its highest setting, putting Neon roughly 51 tokens per second ahead, or about 1.9X faster. That places Neon close to Gemini 3.8 Flash at 116 tokens per second on high. Hydrogen still wins on raw output at 167 on high and 171 on xhigh. Latency is the catch: Priors lists about 57 seconds to the first token for Neon at xhigh. Google's launch post gives no absolute completion times in seconds or minutes for a matched Neon and Carbon workload, so the clearest comparison on that measure comes from a customer. "In our early evaluations, it matched or exceeded Gemini 4 Carbon on overall quality at 3x the speed and 60% of the cost," said Thai Tran, Glean's AI product lead, in Google's announcement.
Vals adds a second, less flattering lens: cost per completed test. Vals lists Sol at $3.18 per test against $16.18 for Sonnet. Sol trails Sonnet by nearly five percentage points, yet the two carry identical standard short-context input and output prices. That gap shows how much a model's behavior during a job, how many tokens it burns thinking, how many retries it takes, changes the real bill. Google's announcement does not give an equivalent Vals dollar figure for Neon, leaving an important part of that three-way comparison open. Reasoning settings also apply only to the Artificial Analysis column, and providers define those settings separately, so two models running at maximum effort can spend very different amounts of time and money on the same question.
For buyers, the practical read is that Google now has an answer for three different budgets: Hydrogen for high-volume work, Neon for harder jobs that recur often, and Carbon for extended projects. The hierarchy only partly predicts the results. At its highest setting, Neon edges past Carbon on the general intelligence index, and on the professional tasks measured by Vals, all three Google models finish within about 2.5 percentage points of one another. That compression is the real story for anyone routing production traffic. If a cheaper tier lands within a couple of points of the expensive one on the tasks you actually run, the routing decision stops being about capability and starts being about unit economics.
The strategic stakes land on procurement calendars. Carbon's discount expires Nov. 30, which gives teams a short window to benchmark Neon against their own workloads before the relative prices shift. Anyone with long-context pipelines should re-run the math on Sol above 272,000 input tokens, where its output-price edge disappears. And anyone paying flagship rates for mid-tier work should note that Neon's claim is not that it is the best model available, but that it is close enough on two engineering benchmarks and the intelligence index to make the premium hard to justify. The mid-tier is where the volume lives, and that is exactly where Google just cut the price of being good enough.
This story's Key Insights and Take-aways are locked.
Create a free account to unlock Executive Actions for one credit.
Register to UnlockAlways free for Executives Club members. Join the Club
More in Technology
Anthropic mathematician claims Claude helped crack the Riemann hypothesis
If the proof holds, AI has toppled math's most famous open problem; if it does not, it is a case study in announcing breakthroughs before anyone verifies them.
AI's $4 Trillion Wipeout Erases Nasdaq's Gains for the Year
A brutal week took roughly $4 trillion off the AI sector and left the Nasdaq flat for the year, and the damage ran the full length of the supply chain.
Anthropic mathematician says AI proved the Riemann hypothesis. Verification will take time.
A single unverified claim from inside Anthropic now sits against more than 160 years of open mathematics, and the gap between asserting a proof and surviving peer review is the entire story.


