Model

Falcon-40B

OPEN

TII · Falcon family · released May 25, 2023

Topped the Hugging Face Open LLM Leaderboard in mid-2023, trained on 1T RefinedWeb tokens; first UAE-built model to lead open-source rankings.

ReasoningCodingVisionFunction callingTool useAgentic

241.4

Elo · rank #357

Parameters

40B

Active params

40B (dense)

Context

2K tokens

Architecture

Causal decoder-only Transformer, 40B (60 layers, multiquery attention, RoPE, FlashAttention)

License

Apache 2.0

Languages

11+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

ARC-ChallengeReasoning61.9%#83

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning81.9%#20

best: Phi-3-medium (14B) · 97.7%

DROPReasoning6.41#71

best: Hunyuan-T1 · 93.1

FIN-benchKnowledge42.6%#6

best: Poro 34B · 66.3%

GSM8KMath21.5%#164

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning85.3%#39

best: Claude 3 Opus · 95.4%

MMLUKnowledge56.9%#216

best: OpenAI o3 · 92.9%

TruthfulQAKnowledge41.6%#75

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning81.3%#31

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

23 GB

VRAM @ FP16

80 GB

Fits on (Q4)

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

Throughput data unavailable.

Quantizations

GGUF · GPTQ · AWQ

Falcon family

Elo progression across releases

Falcon-40BMay 2023241.4 Falcon-7BMay 2023-185.3 Falcon-180BSep 2023402.5

Resources

HFtiiuae/falcon-40b DOCSfalconllm.tii.ae/falcon-40b.html

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