whatmodel.sh / OpenThinker2 32B
OpenThinker2 32B
↗ share
+ contribute
OpenThinker2 32B is an open-weights language model from Open Thoughts, released April 2025. Its published size is 32.76B parameters and a 33k-token context window. Licence: apache-2.0.
Catalogue identity notes
Maker repo open-thoughts/OpenThinker2-32B @0aefafe: licence tag apache-2.0; fine-tune of Qwen/Qwen2.5-32B-Instruct on OpenThoughts2-1M. 32,763,876,352 parameters (BF16) from the Hugging Face safetensors metadata; dense, so active = total. Download 65.53 GB = sum of the 14 safetensors shards at the pinned revision. Released 2025-04: month the maker's repo was created (createdAt 2025-04-03). Distinct from OpenThinker-32B (v1, not catalogued) and OpenThinker3-7B. Ollama's openthinker library has '32b-v2-*' tags, but its readme describes only the v1 model, so no Ollama command is proposed. Why proposed: Lands in both Arena-Hard v2 README runs (21 catalogue models each) and, with the proposed card3 entry, in 4 runs of its own card table with DeepSeek-R1-Distill-Qwen-32B and s1.1-32B: 6 results.
$ how it compares
The LLMs ranked either side of OpenThinker2 32B, out of 828 with a score, then more from Open Thoughts.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 329 |  EXAONE Deep 7.8B | 55±2 | local | ||
| 330 |  Qwen2.5-Math 72B Instruct | 55±2 | local | ||
| 331 | OpenThinker2 32B | 54±1 | local | ||
| 332 |  Falcon-H1 34B Instruct | 54±3 | local | ||
| 333 |  Llama 3.3 70B Instruct | 54±1 | api $0.72/M +local | ||
| more from Open Thoughts | |||||
| 327 | OpenThinker 32B | 55±1 | local | ||
| 404 | OpenThinker3 7B | 50±1 | local | ||
| 428 | OpenThinker2 7B | 49±2 | local | ||
$ what it's good at
chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #261 of 393 * · 24% of its score · 2 tests
Beats Ministral 3 14B Reasoning 2512, Ministral 3 14B here, though they rank higher overall. Trails Flan-T5 XXL, Granite 4.0 H 1B, Granite 4.0 1B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena-Hard v2.0 hard prompts (style control, Gemini-2.5 judge) | 3.2% ✎ | ▃·▃▅▅·▃█▆▂▂▂ beats 0/22 | 20% |
| Arena-Hard v2.0 creative writing (GPT-4.1 + Gemini-2.5 judges) | 23.6% ✎ | ▆▂▂▆▂▂▃█▂▃·▂ beats 6/22 | 4.3% |
math▁▂▃▅▆▅▄█▄▁▁▁ #168 of 398 · 49% of its score · 3 tests
Beats MiniMax-M2.1, Mistral Medium 3.5 (128B), GPT-4.1 here, though they rank higher overall. Trails OpenThinker3 7B, Olmo 3 7B Think here, though they rank lower overall.
other tests1 not in a task · 26% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GPQA Diamond lab | 64.1% ✎ | ▁▁▁▁▂▄▅▅███▄ beats 125/310 | 26% |
$ where to run it
llama.cpp · recommended
version
- 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 OpenThinker2 32B · needs ~19.9 GB · from huggingface.co/lmstudio-communitychat in the terminal
llama cli -hf lmstudio-community/OpenThinker2-32B-GGUF:Q4_K_M - Or start a local server and open the address it prints (http://localhost:9931)
llama serve -hf lmstudio-community/OpenThinker2-32B-GGUF:Q4_K_M
LM Studio
version
- Install LM Studio · Open the app once before using the lms command.
- Run OpenThinker2 32B · needs ~19.9 GB · from huggingface.co/lmstudio-community
- Or download it in the terminal
lms get https://huggingface.co/lmstudio-community/OpenThinker2-32B-GGUF@Q4_K_M -y - Then run lms ls to see its name, and lms chat followed by that name
Ollama
version
- Install Ollama
download it (macOS / Windows) ↗
Linuxcurl -fsSL https://ollama.com/install.sh | sh - Run OpenThinker2 32B · needs ~19.9 GB · from huggingface.co/lmstudio-communitychat in the terminal
ollama run hf.co/lmstudio-community/OpenThinker2-32B-GGUF:Q4_K_M
MLX (Mac)
version
- Install MLX (Mac) · Apple silicon Macs only, text only.with uv
uv tool install mlx-lmorpip install mlx-lm - Run OpenThinker2 32B · needs ~18.4 GB, Mac only · from huggingface.co/alexgusevskichat in the terminal
mlx_lm.chat --model alexgusevski/OpenThinker2-32B-mlx-4Bit - Or start an OpenAI-compatible server at http://localhost:8080/v1
mlx_lm.server --model alexgusevski/OpenThinker2-32B-mlx-4Bit
Other ways to run it
- Run it as published · needs ~65.53 GB, NVIDIA only · from huggingface.co/open-thoughts (official)
vllm serve open-thoughts/OpenThinker2-32B
Official weights and documentation: huggingface.co/open-thoughts ↗
$ how the score was worked out · open-thoughts/OpenThinker2-32… 75%, Arena-Hard-Auto v2.0-Preview… 24%
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 |
|---|---|---|
 open-thoughts/OpenThinker2-32B model card · lab-reported | 4 | 75% |
 Arena-Hard-Auto v2.0-Preview leaderboard (repository README) | 2 | 24% |
| prior · pull toward the middle | 1% | |
| weighted average | = 54.4 |
$ 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