Model

InternLM3-8B-Instruct

OPEN

Shanghai AI Laboratory · InternLM3 family · released Jan 15, 2025

Apache-2.0 hybrid deep-thinking/normal-mode model trained on only 4T tokens, saving over 75% of training cost vs peers.

ReasoningCodingVisionFunction callingTool useAgentic

893.9

Elo · rank #236

Parameters

Active params

8B (dense)

Context

128K tokens

Architecture

Dense transformer decoder; hybrid deep-thinking/normal response modes

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

AIMEMath20.0%#140

best: GPT-5.2 · 100.0%

AlpacaEvalHuman preference51.1%#17

best: Tülu 2+DPO 70B · 95.1%

CMMLUKnowledge83.1%#11

best: Doubao-1.5-Pro · 90.9%

DROPReasoning83.1#20

best: Hunyuan-T1 · 93.1

GPQA DiamondReasoning37.4%#231

best: GPT-5.6 · 94.6%

HellaSwagReasoning91.2%#5

best: Claude 3 Opus · 95.4%

HumanEvalCoding82.3%#49

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning79.3%#92

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding17.8%#165

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

MATH-500Math83.0%#61

best: GPT-5 · 99.4%

MMLU-ProKnowledge57.6%#118

best: Claude Fable 5 · 91.5%

MMLUKnowledge76.6%#117

best: OpenAI o3 · 92.9%

WildBenchHuman preference33.1%#10

best: Mistral Small 3.2 (24B) · 65.3%

Run it locally

VRAM @ Q4

5.5 GB

VRAM @ FP16

16 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF Q4 · GPTQ Int4

InternLM3 family

Elo progression across releases

InternLM3-8B-InstructJan 2025893.9

Resources

HFinternlm/internlm3-8b-instruct GHInternLM/InternLM DOCSinternlm.readthedocs.io

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