Model

Falcon-7B

OPEN

TII · Falcon family · released May 25, 2023

Apache-2.0 causal decoder trained on 1.5T RefinedWeb tokens; the small, widely-adopted entry point of the original Falcon line.

ReasoningCodingVisionFunction callingTool useAgentic

-307.6

Elo · rank #448

Parameters

Active params

7B (dense)

Context

2K tokens

Architecture

Causal decoder-only Transformer, 7B (32 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-ChallengeReasoning44.5%#137

best: Llama 3.1 405B · 96.9%

ARC-EasyReasoning73.6%#38

best: Phi-3-medium (14B) · 97.7%

GSM8KMath4.6%#186

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning76.3%#106

best: Claude 3 Opus · 95.4%

LAMBADAReasoning74.9%#10

best: GPT-3 175B · 86.4%

MMLUKnowledge28.0%#275

best: OpenAI o3 · 92.9%

OpenBookQAReasoning44.6%#38

best: Claude 1 · 90.8%

PIQAReasoning80.3%#40

best: GPT-4o mini · 93.1%

TruthfulQAKnowledge34.3%#98

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

WinoGrandeReasoning67.2%#110

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

6 GB

VRAM @ FP16

16.2 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 · GPTQ · bitsandbytes 4-bit

Fine-tune it

Permissive

QLoRA6.4 GB1× RTX 3060 12GB

LoRA16.6 GB1× RTX 3090 24GB

Full fine-tune114.0 GB1× H200 141GB

QLoRA SFT on ~10k samples ≈ $4.10 (1× RTX 3060 12GB)

Plan a fine-tune →

Falcon family

Elo progression across releases

Falcon-40BMay 2023131.5 Falcon-7BMay 2023-307.6 Falcon-180BSep 2023342.4

Resources

HFtiiuae/falcon-7b DOCSfalconllm.tii.ae

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