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. Highly cost-efficient for large-scale workloads

• Extremely low input cost ($0.30/M) and affordable output cost.
• Built for production environments where throughput and budget matter.
• Significantly cheaper than competitors like o4-mini, Claude Sonnet, and Grok on text workloads.

2. Fast performance optimized for everyday tasks

• Ideal for summarization, chat, extraction, classification, captioning, and lightweight reasoning.
• Designed as a high-speed “workhorse model” for apps that require low latency.

3. Built-in “thinking budget” control

• Adjustable reasoning depth lets developers trade off latency vs. accuracy.
• Enables dynamic cost management for large agent systems.

4. Native multimodality across all major formats

• Inputs: text, images, video, audio, PDFs.
• Outputs: text + native audio synthesis (24 languages with the same voice).
• Great for conversational agents, voice interfaces, multimodal analysis, and captioning.

5. Industry-leading long context window

• 1,000,000 token context window.
• Supports long documents, multi-file processing, large datasets, and long multimedia sequences.
• Stronger MRCR long-context performance vs previous Flash models.

6. Native audio generation and multilingual conversation

• High-quality, expressive audio output with natural prosody.
• Style control for tones, accents, and emotional delivery.
• Noise-aware speech understanding for real-world conditions.

7. Strong benchmark performance for its cost

• 11% on Humanity's Last Exam (no tools) - competitive with Grok and Claude.
• 82.8% on GPQA diamond (science reasoning).
• 72.0% on AIME 2025 single-attempt math.
• Excellent multimodal reasoning (79.7% on MMMU).
• Leading long-context performance in its price tier.

8. Capable coding assistance

• 63.9% on LiveCodeBench (single attempt).
• 61.9%/56.7% on Aider Polyglot (whole/diff).
• Agentic coding support + tool use + function calling.

9. Fully supports tool integration

• Function calling.
• Structured outputs.
• Search-as-a-tool.
• Code execution (via Google Antigravity / Gemini API environments).

10. Production-ready availability

• Available in: Gemini App, Google AI Studio, Gemini API, Vertex AI, Live API.
• General availability (GA) with stable endpoints and documentation.