AI Agent Hub
Back to models
Google: Gemini Flash Latest logo

Google: Gemini Flash Latest

Closed Source ~google Released 2026-09-02
-- 1M context Proprietary

About this model

Google: Gemini Flash Latest is a non-versioned Gemini API alias (gemini-flash-latest) that Google hot-swaps to the current production Flash model. As of September 2026 it resolves to Gemini 3.8 Flash (gemini-3.8-flash), Google’s GA Flash tier aimed at high-volume, low-latency agents, software engineering, and knowledge work rather than maximum Pro-tier capability.

Gemini 3.8 Flash is natively multimodal (text, image, audio, and video inputs; text output) with up to about 1M input tokens and 64K output tokens, configurable thinking levels (low, medium, high), and built-in tool use including function calling, code execution, search grounding, and computer-use previews. It is distributed through the Gemini API, Google AI Studio, the Gemini app, Gemini Enterprise Agent Platform, and agent products such as Google Antigravity.

Relative to earlier Flash releases, Google reports strong gains on long-horizon coding and agent benchmarks (for example DeepSWE v1.1, Terminal-Bench 2.1, and SWE-Bench Pro) while keeping Flash-class pricing. Parameter count and weights are not published; access is API-only with no local deployment requirement.

Benchmark Scores

HLE
45.4
DeepSWE
73.7
GDPval-AA
1545.0
HLE-Verified
54.9
SWE-Bench-Pro
61.6
Terminal-Bench-2.1
90.8

Technical Specs

  • Architecture: Sparse mixture-of-experts Transformer
  • Context Window: 1,048,576 tokens
  • Input Modalities: text, image, audio

Hardware Requirements

  • Compute: API only

Pricing

Input Output Currency
0.75 / 1M tokens 3.75 / 1M tokens USD