Model

Yi-1.5 34B

OPEN

01.AI · Yi-1.5 family · released May 13, 2024

Continually pretrained on 500B extra high-quality tokens; separate 16K/32K-context chat checkpoints also released.

ReasoningCodingVisionFunction callingTool useAgentic

530.7

Elo · rank #295

Parameters

34.4B

Active params

34.4B (dense)

Context

4K tokens

Architecture

Dense decoder-only transformer, 34B

License

Apache 2.0

Languages

—

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AGIEvalReasoning71.1%#4

best: OLMo 3-Think 32B · 88.2%

AlpacaEvalHuman preference36.6%#30

best: Tülu 2+DPO 70B · 95.1%

ARC-ChallengeReasoning66.3%#60

best: Llama 3.1 405B · 96.9%

Arena-HardHuman preference42.6%#45

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

BIG-Bench HardReasoning59.8%#79

best: ERNIE 4.5 300B-A47B · 94.3%

CMMLUKnowledge84.8%#6

best: Doubao-1.5-Pro · 90.9%

GPQA DiamondReasoning36.3%#237

best: GPT-5.6 · 94.6%

GSM8KMath82.7%#72

best: Llama 3.1 405B · 96.8%

HumanEval+Coding40.2%#31

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding46.3%#129

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning31.2%#142

best: Gemma 4 26B A4B · 98.5%

MATH-500Math15.4%#185

best: GPT-5 · 99.4%

MBPP+Coding55.4%#32

best: Llama 3.1 405B · 88.6%

MBPPCoding65.5%#48

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLU-ProKnowledge47.1%#139

best: Claude Fable 5 · 91.5%

MMLUKnowledge77.1%#115

best: OpenAI o3 · 92.9%

MT-BenchHuman preference8.5#27

best: Hunyuan-Large (A52B) · 9.4

Run it locally

VRAM @ Q4

20 GB

VRAM @ FP16

69 GB

Fits on (Q4)

RTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Q4 GGUF (~20GB) fits a 24GB card only with reduced context headroom; no verified llama.cpp 4090 tok/s figure found.

Quantizations

GGUF Q4 · EXL2

Yi-1.5 family

Elo progression across releases

Yi-1.5 34BMay 2024530.7 Yi-1.5 9BMay 2024396.1

Resources

HF01-ai/Yi-1.5-34B GH01-ai/Yi-1.5 APIAvailable via 01.AI API

API price weights · each benchmark row carries its own source badge (see methodology)