Gemini 2.5 Flash Lite
A lightweight model in Google's Gemini series, designed for low-cost, high-throughput scenarios. It offers advantages in speed and pricing, making it suitable for high-frequency tasks such as classification and routing.
Timeline (last 90 days)
In independent comparison tests, Jev 1.13's average cost per call was $0.0000188, Mistral Small 3.2 cost 1.4 times Jev, and Gemini 2.5 Flash Lite cost 1.7 times.
In the same tests, Jev's median response time was 352 ms, Mistral Small 3.2 was 1,343 ms, Gemini 2.5 Flash Lite was 877 ms, and GPT-5 nano was 7,504 ms.
In the initial tests the chat models could not reliably return the declared type because structured output was not enabled; after merely adding response_format: { type: \"json_schema\", strict: true }, Gemini went from 2/5 to 5/5 and GPT-5 nano from 4/5 to 5/5.
On the support-ticket cases with JSON Schema strict mode enabled, Jev took 386 ms, Gemini 896 ms, Mistral 1,672 ms, and GPT-5 nano 7,006 ms.
Vercel previously used Gemini 2.5 Flash Lite for classification work; after switching to Jev, the speed was approximately 6 times faster and evaluation was largely saturated.
All Facts (5)
In independent comparison tests, Jev 1.13's average cost per call was $0.0000188, Mistral Small 3.2 cost 1.4 times Jev, and Gemini 2.5 Flash Lite cost 1.7 times.
50%UnverifiedIn the same tests, Jev's median response time was 352 ms, Mistral Small 3.2 was 1,343 ms, Gemini 2.5 Flash Lite was 877 ms, and GPT-5 nano was 7,504 ms.
50%UnverifiedIn the initial tests the chat models could not reliably return the declared type because structured output was not enabled; after merely adding response_format: { type: \"json_schema\", strict: true }, Gemini went from 2/5 to 5/5 and GPT-5 nano from 4/5 to 5/5.
50%UnverifiedOn the support-ticket cases with JSON Schema strict mode enabled, Jev took 386 ms, Gemini 896 ms, Mistral 1,672 ms, and GPT-5 nano 7,006 ms.
50%UnverifiedVercel previously used Gemini 2.5 Flash Lite for classification work; after switching to Jev, the speed was approximately 6 times faster and evaluation was largely saturated.
50%