benchrPricing breakdown
| Tier | Rate / 1M tokens |
|---|---|
| Standard input | $1.50 |
| Standard output | $9.00 |
| Cached input | $0.15 |
| Free tier | Available (rate-limited) |
| Context window | 1,048,576 tokens |
| Max output | 65,536 tokens |
Latency needs a deployment test
Google does not publish one universal tokens-per-second or time-to-first-token guarantee for this comparison, and benchr has not run a controlled cross-region latency study. The undocumented values previously shown here have been removed.
Benchmark the same prompts, output lengths, settings, and concurrency levels in the deployment region you intend to use. Record first-token time plus p50, p95, and p99 completion time, then size capacity from those observations.
Caching: the published cached-input rate
Cached input lists at $0.15/1M, 90% below the standard input rate. For an illustrative workload with a 90% cache-hit rate, the input-only arithmetic is $0.15 × 0.9 + $1.50 × 0.1 = $0.285 per million. This scenario does not include output, cache-storage, or feature charges and is not a market-wide “cheapest” claim; compare current provider terms for your workload.
Vision: an editorial capability field
Gemini 3.5 Flash supports image, video-frame, audio, and PDF inputs through the Gemini API. benchr's “92” vision field is an editorial capability score, not an independent multimodal benchmark. Test it on the actual screenshots, scans, product images, or video frames your pipeline receives.
Google Search grounding
Google lists Search grounding as a Gemini API capability. Availability, supported regions, request options, and charges can change, so verify the current documentation before designing around it. Grounding can retrieve web results for a response, but its effect on factual accuracy and whether it complements or replaces any retrieval component must be evaluated on your own queries.
The free tier
Google lists a rate-limited free tier for eligible use. Quotas and eligibility can change, so check the current provider page rather than relying on a fixed request-per-minute or daily limit here. It can support prototyping and low-volume evaluation; production suitability depends on current quotas, reliability requirements, and traffic.
Cost scenarios
At 10M input + 3M output per month, standard pricing is $15 + $27 = $42/month. With a hypothetical 90% cache-hit rate on input, the arithmetic is $1.35 + $1.50 = $2.85 input + $27 output = $29.85/month. This is a price scenario only; check the 65K output cap and measure latency separately.
Use-case fit
Test for: Multimodal tasks requiring image, video, audio, or document input; workloads using Search grounding; and pipelines that benefit from the published 1M context window.
Check first: Measure coding quality and latency yourself because Google does not publish a directly comparable SWE-bench Verified or universal latency figure here. Also verify the published output cap and your data-residency requirements.
Decision checklist
Before routing to Gemini 3.5 Flash, verify the listed 65K max-output limit against your longest expected outputs. Measure first-token latency from your deployment region because service location, load, and request shape can change the observed distribution.
If you're evaluating against GPT-5 on cost: run the comparison with caching enabled. Gemini 3.5 Flash's $0.15/1M cached rate makes it materially cheaper than uncached GPT-5 ($1.25/1M) for cache-heavy workloads.