Gemini 3.6 Flash and GPT-5 Mini reached a perfect 100% score on 7-day standing benchmarks in measurements recorded on September 6, 2026. Both models completed the evaluation suite with zero total spend. Meanwhile, latency tests across hourly checks yielded a 100% success rate for all tested endpoints, with [Gemini 2.5 Flash Lite](/models/google-gemini-2-5-flash-lite) posting the fastest overall median response time of 537ms.
Benchmark
Over the last 7 days, six models recorded results on the standing benchmark. Gemini 3.6 Flash and GPT-5 Mini led the field by capturing 100% of available points without incurring any spend. OpenAI's GPT-5.6 Terra placed third, securing 89.9% of available points at $0 spend.
Google's Gemini 3.1 Flash Lite achieved 85.6% of available points, followed by Gemini 2.5 Flash at 79.5%. GPT-5.6 Luna rounded out the evaluated group with a score of 75.2%. Every model tested on the benchmark registered $0 in total spend for the evaluation period.
| Model | 7-Day Benchmark Score | Total Spend |
|---|---|---|
| Gemini 3.6 Flash | 100% | $0 |
| GPT-5 Mini | 100% | $0 |
| GPT-5.6 Terra | 89.9% | $0 |
| Gemini 3.1 Flash Lite | 85.6% | $0 |
| Gemini 2.5 Flash | 79.5% | $0 |
| GPT-5.6 Luna | 75.2% | $0 |
Speed
Latency evaluations conducted over the last 24 hours showed consistent reliability, with 100% of hourly uptime checks succeeding across all 13 monitored endpoints. Response times varied widely depending on model class, ranging from 537ms to 3.58s.
Gemini 2.5 Flash Lite delivered the fastest median response time at 537ms. GPT-5.4 Mini followed closely at 629ms. Gemini 3.1 Flash Lite and Gemini 2.5 Flash recorded nearly identical median speeds at 801ms and 802ms, respectively.
Mid-range response times were dominated by OpenAI's 5.6 series models. GPT-5.6 Luna logged a median latency of 1.07s, while GPT-5.6 Terra posted 1.08s and GPT-5.6 Sol registered 1.18s.
Larger and higher-tier models showed longer latencies. Gemini 3.5 Flash recorded a median response time of 2.12s, slightly ahead of Gemini 3.6 Flash at 2.20s. GPT-5 Nano logged 2.29s, while GPT-5 Mini recorded 2.35s. The slowest endpoints in the 24-hour window were the preview and pro-tier models: Gemini 3.1 Pro (preview) recorded a 3.38s median response time, and Gemini 2.5 Pro registered 3.58s.
Prices
Billed costs per million tokens remained distinct across Google's Gemini tiers. Gemini 2.5 Flash Lite stood as the lowest-cost option measured, priced at $0.100 per million input tokens and $0.400 per million output tokens. Gemini 3.1 Flash Lite measured at $0.250 per million input tokens and $1.50 per million output tokens, while Gemini 2.5 Flash billed at $0.300 per million input tokens and $2.50 per million output tokens.
Higher-tier models showed significantly elevated rates. Gemini 2.5 Pro was billed at $1.25 per million input tokens and $10.00 per million output tokens. Gemini 3.5 Flash measured at $1.50 per million input tokens and $9.00 per million output tokens. The highest pricing was observed on Gemini 3.1 Pro (preview), which billed at $2.00 per million input tokens and $12.00 per million output tokens.
These direct measurements prove specific latency, uptime, pricing, and benchmark scores under observed testing conditions on September 6, 2026. They do not prove real-world application performance, production behavior under unmonitored prompt loads, or future pricing structure changes from model providers.
