Model

InternLM2-20B

OPEN

Shanghai AI Laboratory · InternLM2 family · released Jan 17, 2024

Flagship dense InternLM2 model; near-perfect 200K-token needle-in-haystack retrieval, free commercial weight license.

ReasoningCodingVisionFunction callingTool useAgentic

487.1

Elo · rank #306

Parameters

20B

Active params

20B (dense)

Context

200K tokens

Architecture

Dense transformer decoder (GQA)

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

AGIEvalReasoning50.3%#17

best: OLMo 3-Think 32B · 88.2%

BIG-Bench HardReasoning68.3%#63

best: ERNIE 4.5 300B-A47B · 94.3%

C-EvalKnowledge63.0%#45

best: Qwen3.6-Plus · 93.3%

CMMLUKnowledge65.1%#22

best: Doubao-1.5-Pro · 90.9%

GSM8KMath79.6%#85

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning85.8%#35

best: Claude 3 Opus · 95.4%

HumanEvalCoding67.1%#88

best: Claude Opus 4.5 · 99.4%

MATH-500Math31.9%#160

best: GPT-5 · 99.4%

MBPPCoding65.8%#47

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

MMLUKnowledge66.5%#170

best: OpenAI o3 · 92.9%

RACE-HReasoning87.1%#5

best: Claude 3 Opus · 92.9%

TriviaQAKnowledge53.9%#39

best: Sarvam-1 (2B) · 90.6%

WinoGrandeReasoning74.8%#69

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

12 GB

VRAM @ FP16

40 GB

Fits on (Q4)

RTX 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 · AWQ

InternLM2 family

Elo progression across releases

InternLM2-20BJan 2024487.1

Resources

HFinternlm/internlm2-20b GHInternLM/InternLM DOCSinternlm.readthedocs.io

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