whatmodel.sh / Trinity Mini (26B-A3B)
Trinity Mini (26B-A3B)
↗ share
+ contribute
Trinity Mini (26B-A3B) is an open-weights language model from Arcee AI, released December 2025. Its published size is 26.12B parameters (3B active per token) and a 131k-token context window. Licence: openmdw-1.1.
Catalogue identity notes
arcee-ai/Trinity-Mini @cdb81e8 (created 2025-12-01, licence openmdw-1.1). Arcee AI Trinity Mini MoE reasoning model (Dec 2025; 26B total, ~3B active per the card), sibling of the catalogued Trinity Large models. Independent rows: EuroEval English test split (the separate '(val)' entry is not used).
Model information sources
Release date ↗ · Weights licence ↗
HF safetensors metadata 26,123,974,400 parameters; ~3B active is the maker's description (Trinity Mini 26B-A3B).
$ how it compares
The LLMs ranked either side of Trinity Mini (26B-A3B), out of 1132 with a score, then more from Arcee AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 489 |  Gemini 1.5 Flash (002) | 53±1 | comparison only | ||
| 490 |  Infinity-Instruct-7M-Gen-Llama3_1-70B | 53±2 | local | ||
| 491 | Trinity Mini (26B-A3B) | 53±6 | local | ||
| 492 |  Gemma 3 27B | 53±1 | api $0.10/M +local | ||
| 493 |  Phi-4 14B | 53±1 | api $0.087/M +local | ||
| more from Arcee AI | |||||
| 237 | Trinity-Large-Thinking | 63±2 | api $0.39/M +local | ||
| 312 | Trinity-Large-Preview | 59±2 | local | ||
| 603 | AFM-4.5B | 49±6 | local | ||
$ what it's good at
other tests2 not in a task · 84% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| EuroEval HellaSwag (English), test split, 5-shot generative MC, accuracy | 68.7% ✎ | ▃·▁▁▂▁▂▅█▇▄▁ beats 46/100 | 53% |
| EuroEval Life in the UK (English knowledge MC), test split, 5-shot, accuracy | 91.4% ✎ | ▁▁▁▁▁▁▁▂▄█▄▃ beats 121/146 | 31% |
$ where to run it
llama.cpp · recommended
- 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 | shWindowsirm https://llama.app/install.ps1 | iex - Run Trinity Mini (26B-A3B) · needs ~15.94 GB · from huggingface.co/arcee-ai (official)chat in the terminal
llama cli -hf arcee-ai/Trinity-Mini-GGUF:Q4_K_M - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf arcee-ai/Trinity-Mini-GGUF:Q4_K_M
LM Studio
- Install LM Studio · Open the app once before using the lms command.
- Run Trinity Mini (26B-A3B) · needs ~15.94 GB · from huggingface.co/arcee-ai (official)
- Or download it in the terminal
lms get https://huggingface.co/arcee-ai/Trinity-Mini-GGUF@Q4_K_M -y - Then run lms ls to see its name, and lms chat followed by that name
Ollama
- Install Ollama
download it (macOS / Windows) ↗
Linuxcurl -fsSL https://ollama.com/install.sh | sh - Run Trinity Mini (26B-A3B) · needs ~15.94 GB · from huggingface.co/arcee-ai (official)chat in the terminal
ollama run hf.co/arcee-ai/Trinity-Mini-GGUF:Q4_K_M
Official weights and documentation: huggingface.co/arcee-ai ↗
$ how the score was worked out · EuroEval leaderboards (result… 84%
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
| source | results | share of the score |
|---|---|---|
 EuroEval leaderboards (results collection), English generative leaderboard | 2 | 84% |
| prior · pull toward the middle | 16% | |
| weighted average | = 52.8 |
$ improve this page
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