US EditionGoogle's Powerful New AI Model Can Solve Your Most Complex Problems. If You Can Afford It
AUG 1 – Gemini 2.5 Deep Think mode enhances AI reasoning with parallel thinking and scores 87.6% on coding benchmarks, improving complex problem solving for AI Ultra subscribers.
- On Friday, Google DeepMind is rolling out Gemini 2.5 Deep Think to Google AI Ultra subscribers, offering its most advanced reasoning model in the Gemini app.
- Buoyed by research breakthroughs and feedback from early trusted testers, today’s release shows significant improvement over the May preview at Google I/O 2025.
- Benchmark evaluations indicate Gemini 2.5 Deep Think now scores 87.6% on LiveCodeBench, up from 80.4% in May, and Google claims it scored 34.8% on HLE versus Grok 4’s 25.4%.
- In testing, Deep Think is reported to excel at iterative development and produces more detailed, aesthetically pleasing web development tasks for Ultra subscribers.
- Future API access in the coming weeks will be limited by high computational costs, but it could significantly accelerate AI research and problem-solving, Google claims.
Insights by Ground AI
Does this summary seem wrong?
21 Articles
21 Articles
The new offshoot of Gemini 2.5 is partly clearly at the top of the benchmarks in mathematics, coding and reasoning. New techniques and parallel "thinking" are behind this, but they have their price
·Vienna, Austria
Read Full ArticleCoverage Details
Total News Sources21
Leaning Left3Leaning Right0Center8Last UpdatedBias Distribution73% Center
Bias Distribution
- 73% of the sources are Center
73% Center
L 27%
C 73%
Factuality
To view factuality data please Upgrade to Premium
