CekuraBench

Gemini 3.8 Flash-Lite

Every voiceCompare

5th of 26 voices at reading hard parts right

Hard parts read right: 88.6%

  • Also strong at
  • Low price6th of 25 · $19.03
  • Reads numbers and shorthand rightTied 6th of 26 · 85%

Gemini 3.8 Flash-Lite: 5th of 26 voices at reading hard parts right. Its weakest is non-streaming latency, 22nd of 26.

Last updated Methodology by Dileep Chagam

Gemini 3.8 Flash-Lite at a glance

Where it ranks among all 26 voices, in plain words.

Good at
  • Reads hard parts right5th of 26
  • Low price6th of 25
  • Reads numbers and shorthand rightTied 6th of 26
Not so good at
  • Slow to start talking22nd of 26

Hear Gemini 3.8 Flash-Lite

The first read of each test sentence, unedited.

Gemini 3.8 Flash-Lite

Gemini 3.8 Flash-Lite head to head

Pick any voice to set against it. Every pair has its own page.

Compare
MeasureGemini 3.8 Flash-LiteGPT-4o mini TTS
  1. Non-streaming729 ms688 ms
  2. StreamingCan't streamCan't stream
  3. StoppingNo stop messageNo stop message
  4. Accuracy88.6%82.7%
  5. Naturalness4.344.43
  6. Price / 1M$19.03$25.70
  7. Measures won2 of 62 of 6
The check marks the better value on each measure. Every pair has its own page with the full numbers.Open this pair · 325 pairs in all

Its closest rivals on speed

Gemini 3.8 Flash-Lite FAQ

  • Gemini 3.8 Flash-Lite starts in 729 ms on a typical sentence (median) and 849 ms on the slowest 1 in 10 (p90), 22nd of 26 voices.

  • No. Gemini 3.8 Flash-Lite needs the whole reply before it starts.

  • Gemini 3.8 Flash-Lite has no stop message, so the agent has to close the connection to stop it.

  • Gemini 3.8 Flash-Lite read 88.6% of the hard parts right, as speech-to-text heard it. Its weakest kind is codes and addresses, at 84.5%. This is an estimate, not a listening test.

  • Gemini 3.8 Flash-Lite lists at $19.03 per million characters.