Seed 2.1 Turbo

On 24 June 2026, ByteDance released Seed 2.1 Pro and Seed 2.1 Turbo at its Volcano Engine FORCE conference as a two-model family aimed at agent-driven enterprise work. Seed 2.1 Turbo is a faster, cheaper variant built for high-frequency enterprise scenarios, while Seed 2.1 Pro powers the consumer-facing Doubao 2.1 Pro assistant. Both models are distributed through Volcano Engine cloud and Volcano Ark model service. ByteDance reports Seed 2.1 Pro achieves leading scores on Terminal Bench 2.1, SWE-Pro, and SciCode for code generation, and leads OSWorld, MobileWorld, and MMMU-Pro for agent and multimodal tasks. The company claims Pro outperforms Claude Opus 4.6 on several coding, agent, and multimodal benchmarks while offering total cost of ownership close to 80 percent lower. Seed 2.1 Pro improvements focus on code generation, long-chain agent task execution, and multimodal understanding.

Seed 2.1 Pro, the flagship of ByteDance's Doubao Seed 2.1 family announced at the Volcano Engine FORCE conference on 24 June 2026, is offered through the Volcano Engine API alongside the smaller Seed 2.1 Turbo (model id doubao-seed-2-1-pro-260628). The Seed team describes the line as agent-first, targeting long-horizon Across a ByteDance-published comparison suite, Seed 2.1 Turbo records scores including KINA 46.6%, SuperGPQA 67.4%, BeyondAIMER 88%, Workspace Bench 54.7%, Agent Startup Bench 54%, Terminal Bench 2.1 67.6%, SWE-Atlas 30.6%, MathVision with tools 92.7%, MMMU-Pro with tools 82.2%, WorldVQA 48.6%, and ZEROBench with tools

Seed 2.1 Turbo was released on 24 June 2026 by ByteDance, according to the LLM-Stats model tracker, which lists the model as proprietary with hosted access only and points to the official ByteDance Seed blog post at seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity as the technical report LLM-Stats flags the model's common peer comparisons (Seed 2.1 Turbo vs GLM-5.1, Claude Opus 4.6, GPT-5.2, Qwen3.8 Max, Qwen3.7-Plus, and GPT-5.5 Instant) but shows "score pending" across all of them, indicating no independent benchmark numbers have been ingested yet. The page is therefore most useful as a release-date

Volcano Engine (Volcengine) launched the Doubao-2.1 model family in late June 2026, which includes the flagship Doubao-Seed-2.1-Pro and the lightweight Doubao-Seed-2.1-Turbo, the subject model of this entry. APIs for both variants became available on the Volcano Ark platform, with a separate Coding Plan endpoint and a Pricing for the Doubao 2.1 Pro tier was set at RMB 6 per million input tokens and RMB 30 per million output tokens, dropping to RMB 1.2 per million input tokens under cache-hit conditions; Volcengine stated the effective cost in Coding and Agent scenarios could be lowered to roughly RMB 1.96 per million tokens. At the

ByteDance released Seed 2.1 Turbo on August 10, 2026, extending the Seed 2.1 family first announced at the Volcano Engine FORCE conference in late June. The Turbo variant is positioned as a faster, lower-cost build for high-frequency enterprise workloads, with the family targeting coding engineering, long-chain agent execution, and multimodal understanding. On Volcano Engine it ships under the endpoint doubao-seed-2-1-turbo-260628 alongside the Pro endpoint. Launch materials describe Seed 2.1 as "the next-generation large model designed for the Coding and Agent era," covering complex engineering delivery and large-scale production scenarios. The Turbo tier is priced at half of the Pro tier, with Pro listed at roughly $0.85 per million input tokens and $4.15 per million output tokens. ByteDance highlights improved visual understanding, spatial reasoning, and long-context processing across multiple video understanding benchmarks.