Model
Model explorer

Gemini 3.8 Flash-Lite

CLOSED
Google · Gemini 3.8 family · released Sep 16, 2026

Google's September 16, 2026 high-volume tier — the 3.8 generation's Flash-Lite, at $0.25/$1.50 per Mtok (a third of 3.8 Flash's rate). Keeps the full text/vision/audio/video modality stack, which is the Lite line's real differentiator against Qwen Flash-Next and Claude Haiku 5, both of which arrived the same week. The launch table stays narrow (Google saves its multi-suite treatment for Pro/Flash): four headline rows plus OSWorld-Verified and Video-MME Pro, a benchmark revision Google is now co-citing with ByteDance. Filed as its own row rather than an effort config because Google bills Lite as a distinct model id.

ReasoningCodingVisionFunction callingTool useAgentic
2198.2
Elo · rank #65
Parameters
Undisclosed
Active params
Undisclosed
Context
1M tokens
Architecture
Proprietary cost-optimized variant of Gemini 3.8 Flash (sparse MoE, undisclosed size); 1M-token context, natively multimodal
License
Proprietary
Languages
—
API price (in/out)
$0.25 / $1.5
Modalities
text · vision · audio · video
Benchmark results
Bar shows position within the tracked field; marker = field best
DeepSWECoding52.3%#16
best: Muse Spark 1.3 · 75.4%
GDPval-AAAgents1342#38
best: Claude Fable 5 · 1932
HLE-VerifiedKnowledge38.6%#5
best: Gemini 3.8 Flash · 54.9%
MMMU-ProVision74.8%#35
best: Claude Opus 4.7 · 85.5%
OSWorld-VerifiedAgents76.4%#11
best: Claude Fable 5 · 85.0%
SWE-bench ProCoding57.6%#33
best: Claude Fable 5.1 · 81.2%
Terminal-Bench 2.0Coding58.9%#58
best: Gemini 3.8 Flash · 89.4%
Video-MME ProVision81.5%#3
best: Seed 2.2 Pro · 90.4%
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
$0.25
Output / M tok
$1.5
API price $0.25/$1.5 · each benchmark row carries its own source badge (see methodology)