Doubao-Seed-Code
ByteDance's budget agentic coding model — SWE-Bench-Verified record at roughly 60% below market price.
About
Doubao-Seed-Code is ByteDance Seed's coding-specialized model, launched November 11, 2025 through Volcano Engine. It is a proprietary mixture-of-experts model optimized for agentic programming with large-scale reinforcement learning, carrying a 256K-token context for full-codebase work. ByteDance claims a state-of-the-art record on SWE-Bench Verified, on par with Anthropic's Claude Sonnet, and prices it about 62.7% below the industry average — a first-month promo ran at just ¥9.9 (about $1.30).
The closest thing to an independent check is the TRAE agent submission built on it, which reported 70.6% single-attempt and 78.8% parallel pass rates on SWE-bench Verified — though that evaluation is ByteDance-adjacent, not fully independent. It plugs into veCLI, Cursor, Cline, and ByteDance's Trae IDE, and exposes an Anthropic-compatible API for drop-in migration.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported: SWE-Bench Verified record, on par with Claude Sonnet.
ByteDance-adjacent: TRAE agent on Doubao-Seed-Code — 70.6% single-attempt, 78.8% parallel on SWE-bench Verified (not fully independent).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Launch confirmed by SCMP and multiple outlets with consistent pricing and benchmark claims; model is live on aggregators (ZenMux listing with pricing). Docked for: (1) headline benchmark is provider-claimed; (2) the nearest independent-ish eval is ByteDance-adjacent; (3) proprietary, China-first availability.