Google DeepMind

Gemini 3 Flash Preview

Type
language model
Context
1,049K tokens
Max output
66K
Released
17 December 2025
Knowledge cutoff
January 2025
API string
google/gemini-3-flash-preview

Prices by provider

Provider Input, $ per 1M Output, $ per 1M In our data since
qihang-ai $0.07 $0.43 31 Jul 2026
frogbot $0.5 $3 31 Jul 2026
perplexity-agent $0.5 $3 31 Jul 2026
gmi $0.5 $3 31 Jul 2026
302.AI $0.5 $3 31 Jul 2026
AIHubMix $0.5 $3 31 Jul 2026
Abacus.AI $0.5 $3 31 Jul 2026
CrossModel $0.5 $3 31 Jul 2026
GitHub Copilot $0.5 $3 31 Jul 2026
Google DeepMind $0.5 $3 31 Jul 2026
Google Vertex AI $0.5 $3 31 Jul 2026
Jiekou.AI $0.5 $3 31 Jul 2026
Kilo Code $0.5 $3 31 Jul 2026
LLM Gateway $0.5 $3 31 Jul 2026
Merge Gateway $0.5 $3 31 Jul 2026
NanoGPT $0.5 $3 31 Jul 2026
OpenRouter $0.5 $3 31 Jul 2026
OrcaRouter $0.5 $3 31 Jul 2026
Requesty $0.5 $3 31 Jul 2026
ZenMux $0.5 $3 31 Jul 2026
Venice AI $0.7 $3.75 31 Jul 2026

Only the base tier and only per-token prices are shown. Batch, discounted and cached rates, as well as prices per image or per second of video, do not go into this table: they cannot stand in the same column as a price per million tokens.

The date in the last column is the day this price first entered our collection. The price may well be older: before that day we simply were not recording it. It has not changed since — otherwise a new row with a new date would stand in its place.

Measurement results

Task set Result Run conditions Measured by
Mock AIME 2024–2025 — olympiad problems 92.77 % · with a tuned harness Epoch evaluations
GeoBench — locating a place from a photograph 88 % GeoBench leaderboard
GPQA Diamond — graduate-level questions 77.61 % · with a tuned harness Epoch evaluations
SWE-bench Verified — fixing bugs in repositories 75.41 % · with a tuned harness Epoch evaluations
SimpleQA Verified — factual accuracy 67.4 % Epoch evaluations
Terminal-Bench — working in the command line 64.3 % · with a tuned harness Terminal-Bench v2 Leaderboard
WeirdML — unusual machine learning tasks 61.6 % WeirdML Leaderboard
SimpleBench — trick questions 53.32 % SimpleBench Leaderboard
FrontierMath, levels 1–3 51.23 % Epoch evaluations
BALROG — game environments 48.1 % Balrog Leaderboard
Chess puzzles 34.76 % Epoch evaluations
ARC-AGI-2 33.61 %
APEX-Agents 24 %
ARC-AGI — generalising to unseen patterns 21.5 % · with a tuned harness
FrontierMath, level 4 — research-grade problems 17.07 % Epoch evaluations
GSO-Bench — code optimisation 9.8 % GSO Leaderboard