Gemini 3.8 Flash

Google · September 2, 2026

● activeCloseddecoder onlymultimodal
Context Window2M tokens

Description

Google's ultra-dense agentic workhorse engineered for continuous background tool execution, code refactoring, and multi-modal streaming.

Benchmark Scores

MMLUMassive Multitask Language Understanding — 57 subjects
90.8%
HumanEvalCode generation pass@1 — Python problems
93.6%
MATHMATH benchmark — competition-level problems
94.7%
SWE-benchReal-world software engineering
67.8%

Key Innovations

Agentic
AgenticModels that can autonomously plan, execute multi-step tasks, use tools, and self-correct without human intervention.
Long Context
Long ContextAbility to process very long inputs (100K+ tokens), enabling analysis of entire codebases or books.
Tool Use
Tool UseAbility to call external tools, APIs, and functions — enabling web browsing, code execution, and real-world actions.
Code Gen
Code GenAbility to write, debug, and understand programming code across multiple languages.

Related Research (1)

2026 · arXiv:2609.37725

Introduced Context Language Models (CLMs) which dynamically read, append, and rewrite their active context window as an editable persistent file durin…

Enabled By

Cerebras CS-4 (Nexus)CEREBRAS · August 2026
Nexus Rack-scale Cluster