Gemini 3.5 Flash Is Fast on One Task and Slow on the Next

Gemini 3.5 Flash takes a median of 8.7 seconds to judge whether a claim stays consistent across a document [^1]. Asked to fix broken line breaks in Markdown output, the same model takes a median of 88 seconds [^2]. A latency figure describes a model on one job, not the model in general.

One model, two very different speeds

The claim-consistency median comes from 276 calls, and both tasks were measured under the same 600-second per-call deadline [^1] [^2]. The gap is not an artefact of different time limits.

The two jobs are different kinds of work. One asks the model for a judgment about a claim; the other asks it to repair formatted text. The benchmark reports the medians, not the reasons behind them, so the safe reading is the narrow one: speed measured on one task does not tell you speed on another.

That matters because latency is often treated as a fixed property of a model, quoted once and carried into every design decision. A team that saw the claim-consistency figure and planned a Markdown cleanup step around it would be off by a wide margin.

On the slow job, the newer version is quicker

On Markdown line-break repair, Gemini 3.6 Flash has a median of 53 seconds, against Gemini 3.5 Flash’s 88 [^2]. On that job, moving up a version cuts the typical wait by 35 seconds.

That does not make Gemini 3.6 Flash faster across the board. These figures give no claim-consistency median for it, so whether the newer model keeps its edge on quick judgment work is not measured here.

What to do with it

• Time candidate models on the task you will actually run, not on a headline figure from a different one.
• Treat a version upgrade as a new measurement. A newer model can be faster on one job without that carrying over to the next.
• Check more than speed before switching. These figures include no first-attempt success rate for Gemini 3.6 Flash and no quality comparison on the Markdown task.

The caveat that matters is sample size. The Markdown comparison rests on only 40 synchronous calls for Gemini 3.5 Flash and 77 for Gemini 3.6 Flash, and it reports medians only [^2]. Read the gap between them as a first measurement, not a settled ranking.

To time models on the job you plan to run, open that task in the live benchmark at https://fronset.ai/benchmark/ and compare medians there rather than across tasks. The method is documented at https://fronset.ai/benchmark/methodology/.