whatmodel.sh / Falcon 40B (base)

Falcon 40B (base)

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Falcon 40B (base) is an open-weights language model from TII, released May 2023. Its published size is 41.8B parameters and a 2k-token context window. Licence: apache-2.0.

Catalogue identity notes

Pretrained Falcon-40B (not falcon-40b-instruct).

34.8±1.2#602 of 719 LLMsfrom 13 benchmark results across 5 sources

$ how it compares

The LLMs ranked either side of Falcon 40B (base), out of 828 with a score, then more from TII.

#modelscorewhereRAMreleased
695Tulu 2 DPO 7B Allen Institute for AI35±1local~5.8GB2023-11
696Eurus-2-7B-SFT PRIME-RL35±3local~6.4GB2024-12
697Falcon 40B (base) TII35±1local~27GB2023-05
698StarCoder2 7B BigCode35±1local~6.1GB2024-02
699Granite 3.0 3B-A800M Instruct IBM35±1local~3.7GB2024-10
more from TII
233Falcon-H1R 7B TII60±1local~6GB2026-01
332Falcon-H1 34B Instruct TII54±3local~23GB2025-05
494Falcon-H1 7B Instruct TII46±3local~6GB2025-05

see all LLMs · find the ones that fit your machine

$ what it's good at

coding▁▁▁▃▆██▇▆▄▃▁ #434 of 458 * · 17% of its score · 2 tests

Beats Qwen3.5 2B, Molmo 7B-D (0924), DeepSeek R1 Distill Qwen 1.5B here, though they rank higher overall. Trails Granite 3.0 1B-A400M Instruct, Llama 3.2 1B Instruct, SmolLM2 1.7B Instruct here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
HumanEval24.4% ✎▁▁▁▁▁▃▅▅█▃▂▁ beats 25/190easy · avg 71%▇▇▇▇ 11%sharp · at its level
MBPP36.5% ✎▁▁▁▄▅█▄▅▃▁▂▁ beats 10/81easy · avg 70%▇▇▇ 6.3%at its level
math▁▂▃▅▆▅▄█▄▁▁▁ #372 of 398 * · 7.2% of its score · 1 test

Beats Command R (03-2024), Molmo 7B-D (0924), Mistral 7B Instruct v0.3 here, though they rank higher overall. Trails SmolLM2 1.7B Instruct, LFM2 350M, Gemma 3 1B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
GSM8K22.4% ✎▂▂▃▅▅▆▅▇█▇▄▃ beats 13/146easy · avg 74%▇▇▇ 7.2%sharp · at its level
other tests10 not in a task · 75% of its score
testresultvs the others in its rundifficultycountswhy
Natural Questions (closed book)32.0% ✎▁·▁·▁▂▁▂·▅▂█ beats 27/28medium · avg 58%▇▇▇▇▇▇▇ 18%at its level
BIG-Bench Hard36.2% ✎▁▁▁▁▂▄▅▅█▃▄▁ beats 17/129medium · avg 68%▇▇▇▇▇▇ 16%sharp
ARC Challenge57.7% ✎▆▁▁▂▂▅▇█▄▃▃▂ beats 68/97easy · avg 88%▇▇▇▇ 11%sharp · at its level
Factual knowledge recall, Premium2k exact match (zero-shot)26.6% ✎▅····▃▅·██·█ beats 8/13medium · avg 59%▇▇▇▇ 9.0%at its level
MMLU53.5% ✎▁▁▁▁▁▂▃▅█▃▆▃ beats 47/183easy · avg 80%▇▇▇ 7.6%at its level
Factual knowledge recall, Premium2k exact match (10-shot)42.3% ✎▃▅▃··▅▃·█▃▃▅ beats 9/13easy · avg 67%▇▇▇ 6.4%at its level
C-Eval40.3% ✎▁···▁▁▃▁▂▂▃█ beats 3/33medium · avg 62%▇ 2.8%noisy
Open PT-LLM Leaderboard average (9 Portuguese tasks)60.3% ✎▃▁▁▁▁▁▁▁▃█▅▁ beats 91/276medium · avg 73%▇ 2.3%at its level · noisy
2 more tests
testresultvs the others in its rundifficultycountswhy
TruthfulQA MC241.6% ✎▅▃█▆▁▂▅···▁▁ beats 12/31medium · avg 64%▇ 1.0%blunt · at its level
COPA84.4% ✎▅·▃·▃··█▃··▃ beats 5/8easy · avg 92%▇ 0.7%at its level · noisy

$ where to run it

llama.cpp · recommended

version

  1. Install llama.cpp · Homebrew and winget installs name the commands llama-cli and llama-server instead.
    macOS / Linux
    curl -LsSf https://llama.app/install.sh | sh
    Windows
    irm https://llama.app/install.ps1 | iex

    other ways to install ↗

  2. Run Falcon 40B (base) · needs ~25.5 GB · from huggingface.co/mradermacher
    chat in the terminal
    llama cli -hf mradermacher/falcon-40b-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf mradermacher/falcon-40b-GGUF:Q4_K_M

LM Studio

version

  1. Install LM Studio · Open the app once before using the lms command.

    download it (macOS / Windows / Linux) ↗

  2. Run Falcon 40B (base) · needs ~25.5 GB · from huggingface.co/mradermacher

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/mradermacher/falcon-40b-GGUF@Q4_K_M -y
  4. Then run lms ls to see its name, and lms chat followed by that name

Ollama

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Falcon 40B (base) · needs ~24 GB · from ollama.com
    chat in the terminal
    ollama run falcon:40b-text

Other ways to run it

  1. Run it as published · needs ~83.67 GB · from huggingface.co/tiiuae (official)
    uv run --with torch --with 'transformers<5' --with accelerate python -c "from transformers import pipeline; print(pipeline('text-generation', model='tiiuae/falcon-40b', torch_dtype='auto', device_map='auto')('Once upon a time', max_new_tokens=40)[0]['generated_text'])"

Official weights and documentation: huggingface.co/tiiuae ↗

$ how the score was worked out · MiniCPM: Unveiling the Potent… 62%, Hallucinations Leaderboard re… 19%, Towards a Holistic Evaluation… 15% and 2 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
MiniCPM: Unveiling the Potential of Small Language Models with Scalable Training Strategies (Hu et al.)7▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 62%CC BY 4.0
Hallucinations Leaderboard results (HF dataset hallucinations-leaderboard/results)2▇▇▇▇▇▇ 19%Apache-2.0
Towards a Holistic Evaluation of LLMs on Factual Knowledge Recall (arXiv:2404.16164v1)2▇▇▇▇▇ 15%CC-BY-4.0
Open Portuguese LLM Leaderboard (archived)1▇ 2%Apache-2.0
From Test-Taking to Test-Making: Examining LLM Authoring of Commonsense Assessment Items (arXiv:2410.14897v1)1▇ 1%CC-BY-4.0
prior · pull toward the middle▇ 1%
weighted average= 34.8

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json