whatmodel.sh / Jamba2 Mini

Jamba2 Mini

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

Jamba2 Mini is an open-weights language model from AI21 Labs, released January 2026. Its published size is 52B parameters (12B active per token) and a 256k-token context window. Licence: apache-2.0.

Catalogue identity notes

AI21 docs (observed 2026-10-09): 'jamba-mini currently points to jamba-mini-2-2026-01', 'jamba-mini-2 points to jamba-mini-2-2026-01', Model Details row 'Jamba Mini ... Version 2, Snapshot 2026-01', and the docs' model card text 'Jamba2 Mini blends efficiency and steerability into a 12B-active parameters model'. Weights: ai21labs/AI21-Jamba2-Mini @24cbbd2 (created 2026-01-06, licence apache-2.0). The Vectara leaderboard row 'ai21labs/jamba-mini-2' is the API id jamba-mini-2 (= jamba-mini-2-2026-01). Not Jamba Mini 1.7 or 1.5 (separate entries).

Model information sources

Release date ↗ · Weights licence ↗

AI21 docs Model Details table: 'Jamba Mini 52B parameters (12B active) 256K 2 2026-01'; card: '12B active parameters (52B total)'; HF safetensors metadata 51,570,323,328 (BF16).

51.6±14.9#402 of 949 LLMsfrom 1 benchmark result across 1 source* thinly tested, treat the rank as provisional

$ how it compares

The LLMs ranked either side of Jamba2 Mini, out of 1132 with a score, then more from AI21 Labs.

#modelscorewhereRAMreleased
524Gemma 3 12B Google52±1api $0.075/M +local~9.1GB2025-03
525Qwen2.5 Coder 7B Instruct Alibaba52±1local~6GB2024-09
526Jamba2 Mini AI21 Labs52±15api price unknown +local~32GB2026-01
527Llama-3-8B-Instruct-SkillMix Princeton PLI52±2comparison only2024-09
528Qwen2.5 14B Instruct Alibaba52±1local~10GB2024-09
more from AI21 Labs
332Jamba 1.6 Large AI21 Labs58±11local~839GB2025-02
485Jamba 1.6 Mini AI21 Labs53±10local~34GB2025-02
560Jamba 1.5 Large AI21 Labs50±1local~839GB2024-08

see all LLMs · find the ones that fit your machine

$ what it's good at

other tests1 not in a task · 1.0% of its score
testresultvs the others in its rundifficultycountswhy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)94.7% ✎▁▂▁▂▅█▂▃▄▃▁▁ beats 85/98very easy · avg 90%▇ 1.0%blunt

$ where to run it

Through an API

api · price unknown · 256k context · docs.ai21.com ↗

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 Jamba2 Mini · needs ~31.2 GB · from huggingface.co/bartowski
    chat in the terminal
    llama cli -hf bartowski/ai21labs_AI21-Jamba2-Mini-GGUF:Q4_K_M
  3. Or start a local server and open the address it prints (http://localhost:9931)
    llama serve -hf bartowski/ai21labs_AI21-Jamba2-Mini-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 Jamba2 Mini · needs ~31.2 GB · from huggingface.co/bartowski

    open in LM Studio ↗

  3. Or download it in the terminal
    lms get https://huggingface.co/bartowski/ai21labs_AI21-Jamba2-Mini-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 Jamba2 Mini · needs ~31.2 GB · from huggingface.co/bartowski
    chat in the terminal
    ollama run hf.co/bartowski/ai21labs_AI21-Jamba2-Mini-GGUF:Q4_K_M

MLX (Mac)

version

  1. Install MLX (Mac) · Apple silicon Macs only, text only.
    with uv
    uv tool install mlx-lm
    or
    pip install mlx-lm
  2. Run Jamba2 Mini · needs ~29 GB, Mac only · from huggingface.co/ncls-p
    chat in the terminal
    mlx_lm.chat --model ncls-p/AI21-Jamba2-Mini-mlx-4Bit
  3. Or start an OpenAI-compatible server at http://localhost:8080/v1
    mlx_lm.server --model ncls-p/AI21-Jamba2-Mini-mlx-4Bit

Other ways to run it

  1. Run it as published · needs ~103.14 GB, NVIDIA only · from huggingface.co/ai21labs (official)
    vllm serve ai21labs/AI21-Jamba2-Mini --mamba-ssm-cache-dtype float32 --enable-auto-tool-choice --tool-call-parser hermes --enable-prefix-caching --quantization experts_int8

Official weights and documentation: huggingface.co/ai21labs ↗

$ how the score was worked out · Vectara hallucination leaderb… 1%

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
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)1▇ 1%Apache-2.0
prior · pull toward the middle▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 99%
weighted average= 51.6

$ 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