new Gemini 3 Deep Think, Anthropic $30B @ $380B, GPT-5.3-Codex Spark, MiniMax M2.5

2026-02-12 · Source: AINews · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, extended

Summary

Google DeepMind is rolling out the upgraded Gemini 3 Deep Think V2 reasoning mode to Google AI Ultra subscribers and opening early access to the Vertex AI / Gemini API for select users. Key benchmark achievements include ARC-AGI-2 at 84.6%, Humanity’s Last Exam (HLE) at 48.4% without tools, and a Codeforces Elo of 3455, showcasing Olympiad-level performance in physics and chemistry. The mode emphasizes practical scientific and engineering applications such as error detection in math papers, physical system modeling, semiconductor optimization, and a sketch to CAD/STL pipeline for 3D printing. ARC benchmark creator François Chollet highlights the benchmark's role in advancing test-time adaptation and fluid intelligence, projecting human-AI parity around 2030. This rollout is framed as a productized, compute-heavy test-time mode rather than a lab demo, with cost disclosures for ARC tasks provided.

Key takeaway

For AI Architects evaluating advanced reasoning and coding models, Gemini 3 Deep Think V2 offers unprecedented performance in scientific and engineering domains, including a sketch-to-CAD/STL pipeline. Its high benchmark scores on ARC-AGI-2 and Codeforces Elo suggest it can tackle complex problems efficiently. Consider its Vertex AI / Gemini API early access for integrating advanced reasoning capabilities into your applications, especially for tasks requiring Olympiad-level problem-solving or agentic coding.

Key insights

Advanced AI models like Gemini 3 Deep Think are achieving human-level performance in complex reasoning and coding tasks.

Principles

• Test-time adaptation is crucial for advancing fluid intelligence in AI.
• AI models are becoming cost-effective for high-volume production workloads.

Method

Google's Gemini 3 Deep Think V2 utilizes a compute-heavy, productized reasoning mode for scientific and engineering problem-solving, including a sketch-to-CAD/STL pipeline.

In practice

• Explore Gemini 3 Deep Think for complex scientific and engineering tasks.
• Evaluate MiniMax M2.5 or GLM-5 for cost-effective coding agent workflows.

Topics

• Gemini 3 Deep Think
• AI Code Generation
• Large Language Models
• AI Agent Frameworks
• AI Infrastructure

Code references

• pytorch/ao
• flashinfer-ai/flashinfer-bench
• traceopt-ai/traceml
• VincentKaufmann/noapi-google-search-mcp
• ggml-org/llama.cpp

Best for: Investor, Director of AI/ML, AI Architect, AI Engineer, AI Product Manager, Tech Journalist

Related on AIssential

• Google Deepmind upgrades Gemini 3 Deep Think for complex science and engineering tasks — Gemini 3 Deep Think excels in abstract reasoning, outperforming competitors on logic, academic, and coding benchmarks.
• Google's Gemini 3.5 Flash follows Anthropic and OpenAI in making newer AI models significantly pricier — Newer, more capable AI models like Gemini 3.5 Flash are significantly pricier due to increased token consumption for complex agentic tasks.
• 🗞️ Google releases Gemini 3 Deep Think, tops ARC-AGI 2 Benchmark With 84.6% — Advanced AI models are achieving human-level performance in complex reasoning and code generation through enhanced architectural and operational methods.
• AI Supremacy 2026: Anthropic’s $30B Surge, India’s Orbital Data Centers, and Google’s Gemini 3 — The AI industry is rapidly evolving with massive investments, novel infrastructure, advanced reasoning models, and simplified core implementations.
• Gemini 3 Deep Think: Advancing science, research and engineering — Gemini 3 Deep Think is an advanced AI reasoning mode for complex scientific and engineering problems.
• 😺 Google’s sharpest brain yet? 🧠 — Google's Gemini 3.1 Pro now leads in overall intelligence and factual accuracy, offering a cost-effective, advanced AI model.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AINews.