Google unveils Gemini 4 Neon model with lower token pricing and faster performance
Միայն փաստեր

Google unveils Gemini 4 Neon model with lower token pricing and faster performance

Ամփոփում

Google introduced Gemini 4 Neon, a mid-range AI model that offers faster output and reduced token costs compared with its own Carbon model and competing services.

Google launched Gemini 4 Neon, a new mid-range large language model, on its paid Gemini API, Google AI Studio, Android Studio, the Gemini app and the Antigravity development platform for Pro and Ultra subscribers. The company says Neon delivers comparable or higher quality than the higher-priced Gemini 4 Carbon model while operating at three times the speed and at 60% of the cost, according to early internal evaluations.

Neon is priced at $3 per million input tokens and $15 per million output tokens. While these rates are about 50% higher than Anthropic’s Claude Sonnet 5.5 and OpenAI’s GPT-6.2 Sol, Google reports that Neon outperforms both on the Artificial Analysis Intelligence Index at its highest reasoning setting. Compared with Anthropic’s flagship Opus 5.5, Neon’s rates are 25% lower, despite scoring within roughly one percentage point on two engineering benchmarks.

Google highlights use cases such as tracing unexpected billing charges, identifying software defects, testing research hypotheses and generating editable spreadsheets. In a demonstration, Neon agents reconstructed an image across 100,000 spreadsheet cells using more than 500,000 formulas, producing a result that a colleague could inspect and modify.

Independent users reported that Neon helped teams understand complex software architectures quickly and detect subtle vulnerabilities. Developers also noted its ability to examine prior research, test hypotheses and incorporate feedback during machine-learning experiments.

Performance benchmarks show Neon streaming about 110 tokens per second at its highest setting, roughly 1.9 × faster than Carbon, which streams 59 tokens per second. Neon’s first-token latency is reported at around 57 seconds. On the Vals evaluation, which combines finance, software, legal and tax tasks, Neon placed second overall with a 70.9% score and ranked first on legal research and tax tasks.

Pricing comparisons indicate that Neon’s cost advantage narrows at higher effort settings but remains below the scheduled rates of Carbon after December 1. On coding benchmarks, Neon achieved a 46.2% score on FrontierCode 2.0, close to Opus’s 47.3% result, while other models such as Sol and Sonnet scored lower.

Early reactions on X noted Neon’s lower cost, faster output and improved behavior relative to Carbon. Google’s internal teams also highlighted a shorter wait for the first token. The company plans a specialized cyber-security version of Neon for eligible Fairwind Program participants, reporting high scores on vulnerability-finding and patching tests.

Overall, Google positions Neon between its lightweight Hydrogen model and the higher-capacity Carbon model, offering a blend of speed, cost efficiency and multi-modal input support for enterprise workloads.

Աղբյուր

VentureBeat

Շարունակեք կարդալ — անվճար

Ստեղծեք անվճար հաշիվ՝ նորություններին հետևելու համար։ Քարտ չի պահանջվում։