1. Strongest Agentic Coding Model

• State-of-the-art on Terminal-Bench 2.0 (82.7%), Expert-SWE (73.1%), and SWE-Bench Pro (58.6%), outperforming GPT-5.4 on complex coding tasks.
• Holds context across large systems, reasons through ambiguous failures, and carries changes through surrounding codebases with fewer tokens.

2. Higher Intelligence at GPT-5.4 Latency

• Co-designed, trained, and served on NVIDIA GB200/GB300 NVL72 systems to match GPT-5.4 per-token latency while performing at a significantly higher level.
• Uses fewer tokens to complete the same tasks, making it more efficient as well as more capable.

3. Powerful for Knowledge Work & Computer Use

• Scores 84.9% on GDPval (44 occupations) and 78.7% on OSWorld-Verified for autonomous computer operation.
• Excels at generating documents, spreadsheets, and reports; naturally moves across finding information, using tools, and checking output.

4. Scientific Research Co-Scientist

• Leading performance on GeneBench, BixBench, and FrontierMath; helped discover a new proof about Ramsey numbers verified in Lean.
• Strong enough to meaningfully accelerate progress at the frontiers of biomedical and mathematical research.

1. Breakthrough long-context window up to 1,000,000 tokens

• Can process 1 hour of video, 11 hours of audio, 700k+ words, or 100k+ lines of code in a single prompt.
• Supports advanced retrieval, reasoning, summarization, and cross-document tasks.
• Achieves 99% retrieval accuracy on 1M-token Needle-In-A-Haystack tests.

2. Strong multimodal reasoning across video, audio, images, and text

• Can analyze long videos (e.g., full silent films), track events, infer causality, and identify small details.
• Handles large complex documents like manuals, transcripts, and books.

3. High-performance reasoning and problem solving

• Comparable to Gemini 1.0 Ultra across many benchmarks.
• Excels at code reasoning, multi-step explanations, and large-scale codebase analysis.

4. Advanced code understanding and generation

• Performs problem-solving on codebases exceeding 100,000 lines.
• Capable of cross-file reasoning, debugging guidance, API comprehension, and generating structured code improvements.

5. Efficient Mixture-of-Experts (MoE) architecture

• Activates only relevant expert pathways per input.
• Enables faster training, lower latency, and more efficient serving.
• Dramatically improves scalability and inference speed.

6. Exceptional in-context learning capabilities

• Learns new tasks directly from long prompts without fine-tuning.
• Demonstrated by learning to translate a low-resource language (Kalamang) from a grammar manual.

7. High-fidelity multimodal understanding

• Reads, analyzes, and reasons about long PDFs, code repositories, images, and videos together.
• Enables new classes of applications: legal analysis, scientific review, codebase audits, long-form content generation, etc.

8. Safety and reliability first

• Undergoes extensive ethics, safety testing, and red-teaming.
• Improved representational safety and reduced hallucinations compared to previous generations.

9. Available for developers and enterprises

• Accessible via AI Studio and Vertex AI.
• Supports future pricing tiers for expanded context windows.
• Designed for real enterprise-scale workloads.

10. Widely capable mid-size model

• Positioned between Gemini Pro and Gemini Ultra generations.
• Well-balanced: reasoning, multimodality, long-context, and speed.