Overview
Seed-Evolving is a deep-reasoning Seed-series model designed for coding and agent workloads. Its unified model ID automatically serves updated versions; it is not a fixed-version identifier. (docs.volcengine.com)
Capabilities and uses
It supports complex task orchestration, long-range planning, code generation, tool and function calling, structured output, document understanding, and text, image and video understanding. Suitable uses include software development, research, office productivity and workflow automation. (docs.volcengine.com)
Limits
The current official documentation lists a 1024K-token context window and maximum input, a 256K-token maximum output and reasoning content length, and limits of 500 requests per minute and 1,000,000 tokens per minute. Audio understanding, including audio within video, is not supported. Automatic updates can change model behavior. (docs.volcengine.com)
Release date
The supplied date is August 1, 2026, but it could not be verified as the original release date. The official release announcement lists the model as newly released in June 2026, without a specific day, and records updates in August 2026. A separate Coding Plan announcement lists availability on August 21, 2026; that is not established as the model's original release date. (docs.volcengine.com)
Pricing
Mainland China, Volcengine Ark, standard online inference, for input lengths of 0–1024K tokens: CNY 6.00 per million non-audio input tokens and CNY 30.00 per million output tokens. Non-audio cache hits cost CNY 1.20 per million tokens; cache storage costs CNY 0.017 per million tokens per hour. Actual charges are subject to the platform's order and billing terms. Checked November 16, 2026. (console.volcengine.com)
An official international USD price for this model could not be verified on the BytePlus ModelArk pricing documentation checked on November 16, 2026. (docs.byteplus.com)