Google Gemini 4 Carbon
Google's most capable Gemini 4 reasoning model for the hardest coding, agentic, mathematical and scientific work, with text, image, audio and video input and a shared 1M-token context.
Gemini 4 Carbon is a cloud text model from Google. It is multimodal: alongside a text prompt it accepts images, video, and audio, then replies with generated text. Its context window handles up to 1.05M input tokens and up to 66K output tokens, and it supports adjustable reasoning effort for harder problems. Generation can be tuned with system instructions. It runs through Google, OpenRouter, and Replicate using your own API key, from $2.00 per million input tokens.
- Released
- Nov 2026
- Pricing
- $2.00 / 1M in · $10.00 / 1M out
- Inputs
- Text, Images, Video, Audio
- Context window
- 1.05M in · 66K out
- Reasoning
- Adjustable effort
- Controls
- System prompt
Samples
Generated with Gemini 4 Carbon via Replicate. The same three prompts run against every text model in the catalog, shown verbatim, so the only thing that changes between two models’ answers is the model.
Explain what an API is to someone who has never written code. Do not use a restaurant or waiter analogy. Plain prose only — no headings, bullet points, or markdown formatting. Maximum 50 words.
An API is like a standard electrical outlet, except for software. Your phone's weather app does not own satellites, so it plugs into a weather service's system using an API to fetch the forecast. It allows separate programs to safely share information without knowing how the other works.
A two-person startup is deciding whether to build their own user authentication or use a hosted provider. Give them a clear recommendation, then the single strongest argument against your own recommendation. Plain prose only — no headings, bullet points, or markdown formatting. Maximum 150 words.
You should unequivocally use a hosted authentication provider. As a two-person team, your scarcest resource is time, and building undifferentiated features like password resets, social logins, and session security steals focus from your core product. Furthermore, a single vulnerability in a homemade system can destroy user trust overnight. However, the strongest argument against a hosted provider is the trap of vendor lock-in paired with punitive scaling costs. Most providers charge by monthly active users, meaning sudden growth in a free or low-margin product can trigger astronomical bills long before your revenue catches up. When you try to leave, you will discover that migrating user data and exporting password hashes is notoriously difficult, often forcing global password resets that drive user churn or holding your business hostage to escalating enterprise contracts just as you begin to scale.
Write a briefing for a non-technical executive explaining why their company's spending on AI is hard to predict, and what they can do about it. Cover at least three distinct causes. Plain prose only — no headings, bullet points, or markdown formatting. Write exactly 205 words.
Predicting our artificial intelligence expenses is uniquely challenging compared to traditional software investments. Unlike standard tools with fixed monthly licenses, most modern AI services charge based on actual usage. Every employee question or automated document summary incurs a tiny fee, meaning sudden staff enthusiasm can unexpectedly multiply our monthly bill. Second, the technology evolves so rapidly that teams must constantly experiment. Discovering which tools actually solve business problems requires funding multiple small trials, many of which will inevitably be discarded as better options emerge next month. Third, hidden data preparation expenses frequently ambush budgets. Before any smart tool can work safely, our existing company information must be cleaned, organized, and secured, demanding unanticipated staff time and computing power. Fortunately, we can regain financial control without stifling useful innovation. We should immediately establish a central review group to approve new projects and eliminate duplicate spending across departments. Next, we can set strict automated spending limits with our vendors so runaway usage pauses before breaking the budget. Finally, we must require teams to run brief, limited pilot programs first. Measuring the exact cost of a small test allows us to forecast expenses accurately before rolling tools out to the entire organization, turning wild guesses into calculated investments.
Google and Google DeepMind build the Gemini family of multimodal models, the Imagen and Nano Banana image models, the Lyria music models, and the Veo video models.
deepmind.google ↗Google's flagship reasoning model for complex software engineering, deep mathematical and scientific problem solving and autonomous multi-step agents, with a shared 1M-token context.
Google's most capable Flash model, with clear gains over 3.7 Flash in software engineering, agentic tasks, and multi-step reasoning, plus vision and a 1M-token context window.
Google's most efficient multimodal Flash model: an updated reasoning stack with stronger coding and computer-use quality, at the same speed and scale as the rest of the Flash line.
Google's most capable multimodal model with deep reasoning and 1M token context.
Google's 26B mixture-of-experts model with long context and strong multimodal reasoning.
Google's flagship 31B dense model with 256K context and top-tier text and image understanding.