GOOGLE DEEPMIND
Gemini 3.0 Pro
BUILD VERSION
v3.0.2510-rc
ARCHITECTURE
MoE-X Hybrid
Click to Analyze ✨
CONTEXT WINDOW
10,000,000
Click to Analyze ✨
MAX OUTPUT
128k Tokens
Click to Analyze ✨
LATENCY (TTFT)
~40ms
Click to Analyze ✨
Core Capabilities
System 2 Reasoning Engine
NEWDynamically allocates compute for "thinking time" before generating tokens. Capable of self-correction in math, logic proofs, and complex planning tasks without external prompting.
Full-Stack Code Agent
UPDATEDIntegrated Sandbox 2.0 environment allowing the model to write, execute, debug, and deploy code. Surpasses human expert level on SWE-bench.
Try Live Demo ✨
Interact with Gemini 3.0 Preview
OPEN_TERMINAL >
Model Matrix
Gemini Ultra
SOTA
For complex reasoning & R&D.
Size: 1.8T (MoE)
Gemini Pro
BEST VALUE
Balanced for scaling apps.
Size: Unknown (MoE)
Gemini Flash
FAST
Sub-20ms latency tasks.
Size: 8B (Dense)
Performance Benchmarks
MMLU-Pro (Reasoning)
92.5%
MATH (Hard Math)
84.2%
HumanEval (Code)
96.8%
Availability: Private Preview (Oct 2025)
Google AI Studio