AI Potluck
Model components / Fine-tuned / chat models

Doubao Seed 2.0

ByteDance Seed / Volcano Engine

ByteDance's flagship closed-weight instruction-tuned model line, released February 14, 2026 in four variants (Pro, Lite, Mini, Code) via the Volcano Engine API. The Pro tier ($0.47 input / $2.37 output per million tokens) posts frontier-tier benchmarks (98.3 AIME 2025, 88.9 GPQA Diamond, 76.5 SWE-Bench Verified, 3020 Codeforces) at 3.7×–10× lower cost than GPT-5.2 / Claude Opus 4.5. Differentiates on (a) Chinese-language strength and Volcano Engine's mainland-China-native infrastructure, (b) aggressive price-to-performance positioning, and (c) deep distribution via the Doubao consumer chat app, which crossed 100M DAU in early 2026 and reached ~345M MAU by March 2026, the largest consumer AI chat product in China. Doubao Seed 1.6 (October 2025, 256K context) remains in the lineup as the budget tier; ByteDance ships 19 active Doubao API SKUs in 2026.

Doubao Seed 2.0 (ByteDance), foundation model family released 14 Feb 2026; variants Pro / Lite / Mini / Code. Multimodal reasoning/agentic instruct model; proprietary/API-only via Volcano Engine (Volcengine) and the consumer Doubao app. Scored facts are for Seed 2.0 Pro (the flagship). Verified via TechNode coverage of ByteDance's launch June 2026.

Openness

1 high confidence
1.0
weights
closed
data
closed
code
closed
license
Proprietary(Doubao app + Volcano Engine API only

No weights released, served only through the Doubao app and Volcano Engine API. Fully closed.

Adoption

5 medium confidence
5.0

Powers the Doubao app, China's #1 AI chatbot with ~155M weekly active users; Seed 2.0 is the model behind that surface plus Volcano Engine enterprise API. Far above >10M. Attributed to the Doubao app surface the model powers, not a standalone Seed 2.0 figure. WAU sourced from launch coverage; vendor primary page (seed.bytedance.com) returned no figure this run.

Capability

5 low confidence
5.0

Vendor-reported (ByteDance) frontier-competitive math/coding/multimodal scores at ~10x lower token cost; figures are not independently audited. Held with low confidence as ByteDance-claimed; corroborated only via launch coverage, not a vendor benchmark page or independent leaderboard fetched this run.

Unchanged since 2026-07-29 (last edited, not re-checked)