Doubao Seed 2.0

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.

• https://technode.com/2026/02/14/bytedance-releases-doubao-seed-2-0-positions-pro-model-against-gpt-5-2-and-gemini-3-pro/ recorded 2026-08-13

  "The model is now available on the Doubao app and through Volcano Engine API services"; no weight release, corpus or training code mentioned

Powers the Doubao app, China's largest consumer AI chat product, with Seed 2.0 the model behind that surface plus the Volcano Engine enterprise API - far above the >10M threshold. The band is taken on the surface the model powers rather than on a standalone Seed 2.0 figure, and that is said here deliberately: banding on a surface is fair as long as the record names which surface. The figure comes from Caixin's launch-day report, which attributes to Quest Mobile that Doubao led with 155 million weekly active users in December 2025; the TechNode launch piece carries no user count and is kept only for the availability fact. The sibling Doubao record independently bands the same app at 382M monthly active users on the QuestMobile 2026 H1 report, so the two reads agree on the order of magnitude and both sit far above >10M. Confidence stays medium because Quest Mobile is read through press coverage rather than through the report itself.

• https://www.caixinglobal.com/2026-02-15/bytedance-unveils-doubao-20-ai-model-to-tackle-complex-tasks-102414865.html recorded 2026-08-14

  Caixin Global on the Seed 2.0 launch, carrying the user figure the record bands on - Doubao had 155 million weekly active users in December 2025 per Quest Mobile, ahead of DeepSeek and well ahead of Tencent Yuanbao at 20.8 million. It also notes Doubao slipped to third in the iOS free download chart over Lunar New Year 2026.

• https://technode.com/2026/02/14/bytedance-releases-doubao-seed-2-0-positions-pro-model-against-gpt-5-2-and-gemini-3-pro/ recorded 2026-08-13

  Doubao Seed 2.0 launch coverage - available on the Doubao app and through Volcano Engine API services, which is the link between this model and the surface it bands on. It carries no user count and no #1 AI chatbot claim.

ByteDance reports frontier-competitive math, coding and multimodal results that no independent evaluation has audited, so the band is 5 and the confidence low. Its launch post claims gold-medal results in ICPC, IMO and CMO, a higher SuperGPQA score than GPT-5.2, science-field performance on par with Gemini 3 Pro and GPT-5.2, leading video-benchmark results and above-human EgoTempo. The widely circulated AIME 2025 98.3, Codeforces 3020 and VideoMME 89.5 figures are not carried here: they appear on none of the reachable sources - not the TechNode piece, not ByteDance's own Seed 2.0 launch post, not the Volcano Engine docs, not Caixin and not the Wikipedia article - and are traceable only to third-party review blogs that publish no working. What is left is a vendor claim with no independent audit and no leaderboard placement behind it.

• https://seed.bytedance.com/en/blog/seed2-0-%E6%AD%A3%E5%BC%8F%E5%8F%91%E5%B8%83 recorded 2026-08-14

  ByteDance's own Seed 2.0 launch post. "Seed2.0 Pro achieved gold medal results in ICPC, IMO, and CMO tests"; "Seed2.0 Pro has scored higher than GPT-5.2 on SuperGPQA"; overall performance in the scientific field is "on par with Gemini 3 Pro and GPT-5.2"; on video it "ranks among the leading players on key benchmarks, including TVBench, TempCompass, and MotionBench, and surpasses human-level performance on EgoTempo". Availability - "The Seed2.0 Pro and Code models have been launched on the Doubao App and TRAE, and the model API for the Seed2.0 full series is available on Volcano Engine." The post carries no AIME, Codeforces or VideoMME figure.

• https://technode.com/2026/02/14/bytedance-releases-doubao-seed-2-0-positions-pro-model-against-gpt-5-2-and-gemini-3-pro/ recorded 2026-08-13

  "ByteDance said Doubao 2.0 Pro achieved leading results across multiple multimodal, math, and coding benchmarks ... performance comparable to leading industry models, while token pricing is about one order of magnitude lower"; the Pro variant "is benchmarked against GPT 5.2 and Gemini 3 Pro". No individual benchmark figures on the page.