Model

Yi-Lightning

CLOSED

01.AI · Yi-Lightning family · released Oct 16, 2024

API-only MoE flagship; reached #6 on LMArena/Chatbot Arena at launch, first Chinese model to surpass GPT-4o there, at $0.14/M tokens.

ReasoningCodingVisionFunction callingTool useAgentic

1063.8

Elo · rank #201

Parameters

Undisclosed

Active params

Undisclosed

Context

16K tokens

Architecture

Mixture-of-Experts transformer with fine-grained expert segmentation & routing, hybrid attention, optimized KV-caching

License

Proprietary (closed weights, API-only)

Languages

—

API price (in/out)

$0.14 / $0.14

Modalities

text

Benchmark results

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

Arena-HardHuman preference91.8%#8

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

Arena EloHuman preference1287#35

best: Claude Fable 5 · 1505

GPQA DiamondReasoning50.9%#184

best: GPT-5.6 · 94.6%

HumanEvalCoding83.5%#46

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning81.9%#86

best: Gemma 4 26B A4B · 98.5%

MATH-500Math76.4%#76

best: GPT-5 · 99.4%

MT-BenchHuman preference8.75#18

best: Hunyuan-Large (A52B) · 9.4

WildBenchHuman preference65.1%#2

best: Mistral Small 3.2 (24B) · 65.3%

Run it locally

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

Input / M tok

$0.14

Output / M tok

$0.14

Yi-Lightning family

Elo progression across releases

Yi-LightningOct 20241063.8

Resources

DOCSplatform.lingyiwanwu.com/docs APIAvailable via 01.AI API

API price $0.14/$0.14 · each benchmark row carries its own source badge (see methodology)