Runway Praxis-1
An open-weight robot control model that combines video prediction and action generation for adaptable robotics policies.
A multimodal reasoning model for sustained coding agents, knowledge work and scientific applications through Google's paid developer platform.
Visit websiteGemini 4 Carbon is a reasoning model available through the paid Gemini API and Google AI Studio. Google positions it for sustained software engineering, multimodal knowledge work, scientific reasoning and cybersecurity defense.
The stable model ID is gemini-4-carbon. It accepts text, images, video, audio and PDF inputs and produces text, allowing developers to combine several source formats within one reasoning workflow.
Choose low, medium or high reasoning effort to balance token use and latency. Medium is the recommended default.
Input and output share a 1,048,576-token context window.
Background execution can generate beyond the default synchronous output limit, using the remaining shared context budget.
Supports function calling, structured output, code execution, computer use and file search.
Supports grounding with Google Search and Google Maps, with tool pricing separate from model token usage.
Evaluate an agent workflow for analyzing code, planning revisions and checking proposed implementations.
Ask questions across supplied documents, images and recordings.
Compare reasoning approaches and request structured explanations of supplied material.
Usage-based token billing with no Carbon free tier. Standard introductory rates apply through November 30, 2026 and increase on December 1. Batch, Flex, Priority, caching and tools have separate rates.
Output pricing includes thinking tokens. Applies to Standard inference through November 30, 2026; caching and tool charges are separate.
Output pricing includes thinking tokens. Applies to Standard inference starting December 1, 2026; caching and tool charges are separate.
Google recommends medium for most tasks. Evaluate all three levels and choose the lowest effort that meets the workload's quality needs.
No. The configured output budget covers thinking and the final answer together, so reasoning can consume space otherwise available for the response.
Yes. The model supports all three options, with separate pricing from Standard inference.