Model

Doubao-Seed-1.6

CLOSED

ByteDance · Doubao family · released Jun 11, 2025

Adaptive-CoT multimodal flagship, 256K context, ~63% cheaper than Doubao 1.5; spans standard/thinking/flash variants.

ReasoningCodingVisionFunction callingTool useAgentic

1639.9

Elo · rank #122

Parameters

230B

Active params

23B (MoE)

Context

256K tokens

Architecture

Mixture-of-Experts, 230B total / 23B active, with Adaptive-CoT (AdaCoT)

License

Proprietary

Languages

—

API price (in/out)

$0.111 / $0.278

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath86.3%#64

best: GPT-5.2 · 100.0%

Gaokao MathMath96.0%#1

best: this model · 96.0%

GPQA DiamondReasoning81.5%#80

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning13.9%#91

best: Claude Opus 5.5 · 67.7%

LiveCodeBenchCoding66.8%#82

best: DeepSeek-V4-Pro (Think Max) · 93.5%

SWE-bench VerifiedCoding41.8%#121

best: Claude Opus 5 · 97.0%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$0.111

Output / M tok

$0.278

Doubao family

Elo progression across releases

Doubao (1st-gen)May 20241862.8 Doubao-1.5-ProJan 20251382.3 Doubao-Seed-1.6Jun 20251639.9

Resources

DOCSseed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6 APIAvailable via ByteDance API

API price $0.111/$0.278 · each benchmark row carries its own source badge (see methodology)