ByteDance Launches Doubao 2.0 — Frontier-Level AI at One-Tenth the Cost
Overview
ByteDance unveiled Doubao 2.0 on February 15, 2026, timed strategically before China's Lunar New Year. The release is widely seen as a pre-emptive move against competitors like DeepSeek, which generated global headlines and a trillion-dollar tech stock selloff with its Spring Festival launch in 2025.
Benchmark Performance
Doubao 2.0 Pro posts the following results according to ByteDance:
- AIME 2025: 98.3
- Codeforces rating: 3,020
- VideoMME (long-video understanding): 89.5%
The company claims the model matches OpenAI's GPT-5.2 and Google's Gemini 3 Pro on core tasks while costing roughly 10 times less through its Volcano Engine API platform.
Designed for the Agent Era
Doubao 2.0 is explicitly built for agentic use cases — executing complex, multi-step real-world tasks rather than answering isolated questions. The Pro variant is accessible via a new Expert Mode in Doubao's app, desktop client, and web interface.
Competitive Landscape
Doubao holds first place in China's AI chatbot market with 155 million weekly active users, well ahead of DeepSeek's 81.6 million. Alibaba's Qwen has surged from 7 million to 58 million daily active users after a promotional campaign, intensifying the domestic race.
Source: The Jakarta Post
Related Articles
Google is steering Gemini toward cost-controlled production agents rather than a single flagship race. The new 3.6 Flash cuts output token use by 17% versus 3.5 Flash, while 3.5 Flash-Lite reaches 350 output tokens per second.
Inception Labs has released Mercury 2, the first production-ready diffusion language model for reasoning. Running at over 1,000 tokens per second on Blackwell GPUs, it is dramatically faster and cheaper than leading autoregressive competitors.
Shared in LocalLLaMA, autoresearch is a minimal framework where an agent edits PyTorch training code, runs fixed five-minute experiments, and keeps changes that improve validation bits-per-byte.