DeepScaleRAgentica Project
|
Gemini 2.5 Deep ThinkGoogle
|
|||||
Related Products
|
||||||
About
DeepScaleR is a 1.5-billion-parameter language model fine-tuned from DeepSeek-R1-Distilled-Qwen-1.5B using distributed reinforcement learning and a novel iterative context-lengthening strategy that gradually increases its context window from 8K to 24K tokens during training. It was trained on ~40,000 carefully curated mathematical problems drawn from competition-level datasets like AIME (1984–2023), AMC (pre-2023), Omni-MATH, and STILL. DeepScaleR achieves 43.1% accuracy on AIME 2024, a roughly 14.3 percentage point boost over the base model, and surpasses the performance of the proprietary O1-Preview model despite its much smaller size. It also posts strong results on a suite of math benchmarks (e.g., MATH-500, AMC 2023, Minerva Math, OlympiadBench), demonstrating that small, efficient models tuned with RL can match or exceed larger baselines on reasoning tasks.
|
About
Gemini 2.5 Deep Think is an enhanced reasoning mode within the Gemini 2.5 family that uses extended, parallel thinking and novel reinforcement learning techniques to tackle complex, multi-step problems in areas like math, coding, science, and strategic planning by generating and evaluating multiple lines of thought before responding, producing more detailed, creative, and accurate answers with support for longer replies and built-in tool integration (e.g., code execution and web search). Its performance shows state-of-the-art results on rigorous benchmarks, including LiveCodeBench V6 and Humanity’s Last Exam, and it demonstrates notable gains over previous versions in challenging domains, with internal evaluations also indicating improved content safety and tone-objectivity, though with a higher tendency to decline benign requests; Google is conducting frontier safety evaluations and implementing mitigations to manage risks as the model’s capabilities advance.
|
|||||
Platforms Supported
Windows
Mac
Linux
Cloud
On-Premises
iPhone
iPad
Android
Chromebook
|
Platforms Supported
Windows
Mac
Linux
Cloud
On-Premises
iPhone
iPad
Android
Chromebook
|
|||||
Audience
Researchers, students, and developers interested in an AI model capable of mathematical reasoning and logic tasks without requiring heavy hardware
|
Audience
Advanced knowledge workers and technical teams wanting a tool to solve intricate, creative, and quantitative problems with contextual depth and tool-enabled execution
|
|||||
Support
Phone Support
24/7 Live Support
Online
|
Support
Phone Support
24/7 Live Support
Online
|
|||||
API
Offers API
|
API
Offers API
|
|||||
Screenshots and Videos |
Screenshots and Videos |
|||||
Pricing
Free
Free Version
Free Trial
|
Pricing
No information available.
Free Version
Free Trial
|
|||||
Reviews/
|
Reviews/
|
|||||
Training
Documentation
Webinars
Live Online
In Person
|
Training
Documentation
Webinars
Live Online
In Person
|
|||||
Company InformationAgentica Project
Founded: 2025
United States
agentica-project.com
|
Company InformationGoogle
Founded: 1998
United States
blog.google/products/gemini/gemini-2-5-deep-think/
|
|||||
Alternatives |
Alternatives |
|||||
|
|
|
|||||
|
|
|
|||||
|
|
||||||
|
|
|
|||||
Categories |
Categories |
|||||
Integrations
CodeMender
Gemini
Gemini 2.5 Flash
Gemini 2.5 Pro
Gemini 3 Deep Think
Gemini Enterprise
Google AI Ultra
|
Integrations
CodeMender
Gemini
Gemini 2.5 Flash
Gemini 2.5 Pro
Gemini 3 Deep Think
Gemini Enterprise
Google AI Ultra
|
|||||
|
|
|