elyza

ELYZA-Thinking-1.0-llm-jp-4-33b

README

License: apache-2.0

Highlights

ELYZA-Thinking-1.0-llm-jp-4-33b is a reasoning model by ELYZA, Inc., mid-trained and post-trained on llm-jp/llm-jp-4-33b-base for Japanese and English language understanding and generation.

Fully domestic foundation. Built on llm-jp-4, pre-trained from scratch in Japan by LLM-jp; all further training was done by ELYZA.
Focus on Japanese reasoning and instruction following. We localized high-quality math, coding and STEM reasoning data into Japanese and built Japanese instruction-following data with multiple checkable constraints, to improve Japanese reasoning and instruction following.
Specialized in Japanese knowledge recall. We put particular care into Japanese-specific knowledge by synthesizing knowledge data from Japanese Wikipedia and Wikidata and training on it, to improve the model's understanding and recall of Japanese knowledge.

Average performance

Benchmarks are first averaged within 7 capability groups, then across the groups (Japanese / English: that language's benchmarks only).

MoE models

Average performance: MoE models

Dense models

Average performance: dense models

Benchmark Results

Per-benchmark scores of the dense models above. For the MoE models, see the model card of elyza/ELYZA-Thinking-1.0-llm-jp-4-32b-a3b.

Model Overview

Three training stages on top of llm-jp/llm-jp-4-33b-base, all done by ELYZA:

Mid-training: a broader foundation for knowledge, reasoning and agentic use.

Math, coding and STEM problems were translated into Japanese and given newly generated reasoning; the original English data was kept as well.
Agentic data was localized into Japanese while keeping tool calls, function names, arguments and JSON intact.

SFT: the answer format and Japanese ability.

Several million synthesized SFT examples with thinking traces, about half of them high-quality Japanese data; answers were generated by a teacher model and kept only when a program verified them (math equivalence, unit tests, rule checkers).
Complex instructions with multiple checkable constraints, including Japanese-specific ones such as furigana and kanji numerals.
Japanese knowledge data synthesized from the latest Japanese Wikipedia and Wikidata.

Reinforcement learning (RLVR): accuracy on hard problems that imitation alone does not reach.

Verifiable problems across math, coding, Japanese knowledge, complex instruction following and tool use, trained together in one run.
Rewards are computed by programs: answer equivalence, test execution, constraint checkers, and call matching or the final environment state for tool use.
Problems were selected so the model sometimes, but not reliably, solves them; reward loopholes (answers that score without solving) were found and closed before training.

See also the sibling model elyza/ELYZA-Thinking-1.0-llm-jp-4-32b-a3b.

Quickstart

vLLM (recommended)

Install vLLM (tested with 0.30.0) and FlashInfer's prebuilt kernels, so no local CUDA toolkit is needed:

bash

pip install vllm==0.30.0pip install flashinfer-jit-cache==0.6.18.post1 --extra-index-url https://flashinfer.ai/whl/cu130

Download the plugins:

bash

hf download elyza/ELYZA-Thinking-1.0-llm-jp-4-33b --include "vllm_plugins/*" --local-dir .

Start the server:

bash

vllm serve elyza/ELYZA-Thinking-1.0-llm-jp-4-33b \  --max-model-len 65536 \  --trust-remote-code \  --reasoning-parser-plugin vllm_plugins/multi_parser_loader.py \  --reasoning-parser llmjp4 \  --tool-parser-plugin vllm_plugins/llmjp_harmony_tool_parser.py \  --tool-call-parser llmjp_harmony \  --enable-auto-tool-choice

Send a request. The answer is returned in content, and the reasoning separately in the message's reasoning field:

bash

curl http://localhost:8000/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b",    "messages": [{"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}],    "temperature": 0.6,    "top_p": 0.95  }'

Tool calling uses the standard OpenAI tools field; calls come back in tool_calls. The bundled tool parser sets the decoding options it needs (keeping the <|call|> marker), so no extra request parameters are required:

bash

curl http://localhost:8000/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b",    "messages": [{"role": "user", "content": "東京の天気を教えて"}],    "tools": [{      "type": "function",      "function": {        "name": "get_weather",        "description": "Get the current weather of a city",        "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}      }    }]  }'

Transformers

Works with the latest Transformers (tested with 4.57.6 and 5.18.0):

bash

pip install -U transformers accelerate torch

python

import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM model_name = "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b" tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(    model_name,    device_map="auto",    dtype=torch.bfloat16,    trust_remote_code=True,)model.eval() messages = [{"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}]inputs = tokenizer.apply_chat_template(    messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device) with torch.no_grad():    output_ids = model.generate(        **inputs,        max_new_tokens=16384,  # thinking model: leave room for the reasoning before the answer        do_sample=True,        temperature=0.6,        top_p=0.95,    )[0][inputs["input_ids"].shape[-1]:] for m in tokenizer.parse_harmony_message(output_ids.tolist()):    print(f"[{tokenizer.decode(m.channel.token_ids)}]", tokenizer.decode(m.content.token_ids))

💡 Tip Recommended sampling: temperature=0.6, top_p=0.95 (default in generation_config.json).

Acknowledgements

We thank LLM-jp, led by the Research and Development Center for Large Language Models at the National Institute of Informatics, for developing the llm-jp-4 models from scratch in Japan and releasing them under a generous open license. This model is built on llm-jp/llm-jp-4-33b-base, and part of our SFT data comes from llm-jp/llm-jp-4-thinking-sft-data.

How to Cite

text

@misc{elyza2026thinking,    title={elyza/ELYZA-Thinking-1.0-llm-jp-4-33b},    url={https://huggingface.co/elyza/ELYZA-Thinking-1.0-llm-jp-4-33b},    author={Xin Zhao, Congda Ma, Satoshi Tohda, Tasavat Trisitichoke, Akira Sasaki, Masato Hirakawa, Kota Kakiuchi},    year={2026},}

Citations

text

@misc{llm-jp-4-33b-base,    title = {llm-jp/llm-jp-4-33b-base},    url = {https://huggingface.co/llm-jp/llm-jp-4-33b-base},    author = {Research and Development Center for Large Language Models},    year = {2026}}

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

elyza

Model Tree

Base

llm-jp/llm-jp-4-33b-base

Fine-tuned

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Highlights

ELYZA-Thinking-1.0-llm-jp-4-33b is a reasoning model by ELYZA, Inc., mid-trained and post-trained on llm-jp/llm-jp-4-33b-base for Japanese and English language understanding and generation.

Fully domestic foundation. Built on llm-jp-4, pre-trained from scratch in Japan by LLM-jp; all further training was done by ELYZA.
Focus on Japanese reasoning and instruction following. We localized high-quality math, coding and STEM reasoning data into Japanese and built Japanese instruction-following data with multiple checkable constraints, to improve Japanese reasoning and instruction following.
Specialized in Japanese knowledge recall. We put particular care into Japanese-specific knowledge by synthesizing knowledge data from Japanese Wikipedia and Wikidata and training on it, to improve the model's understanding and recall of Japanese knowledge.

Average performance

Benchmarks are first averaged within 7 capability groups, then across the groups (Japanese / English: that language's benchmarks only).

MoE models

Average performance: MoE models

Dense models

Average performance: dense models

Benchmark Results

Per-benchmark scores of the dense models above. For the MoE models, see the model card of elyza/ELYZA-Thinking-1.0-llm-jp-4-32b-a3b.

Model Overview

Three training stages on top of llm-jp/llm-jp-4-33b-base, all done by ELYZA:

Mid-training: a broader foundation for knowledge, reasoning and agentic use.

Math, coding and STEM problems were translated into Japanese and given newly generated reasoning; the original English data was kept as well.
Agentic data was localized into Japanese while keeping tool calls, function names, arguments and JSON intact.

SFT: the answer format and Japanese ability.

Several million synthesized SFT examples with thinking traces, about half of them high-quality Japanese data; answers were generated by a teacher model and kept only when a program verified them (math equivalence, unit tests, rule checkers).
Complex instructions with multiple checkable constraints, including Japanese-specific ones such as furigana and kanji numerals.
Japanese knowledge data synthesized from the latest Japanese Wikipedia and Wikidata.

Reinforcement learning (RLVR): accuracy on hard problems that imitation alone does not reach.

Verifiable problems across math, coding, Japanese knowledge, complex instruction following and tool use, trained together in one run.
Rewards are computed by programs: answer equivalence, test execution, constraint checkers, and call matching or the final environment state for tool use.
Problems were selected so the model sometimes, but not reliably, solves them; reward loopholes (answers that score without solving) were found and closed before training.

See also the sibling model elyza/ELYZA-Thinking-1.0-llm-jp-4-32b-a3b.

Quickstart

vLLM (recommended)

Install vLLM (tested with 0.30.0) and FlashInfer's prebuilt kernels, so no local CUDA toolkit is needed:

bash

pip install vllm==0.30.0pip install flashinfer-jit-cache==0.6.18.post1 --extra-index-url https://flashinfer.ai/whl/cu130

pip install vllm==0.30.0pip install flashinfer-jit-cache==0.6.18.post1 --extra-index-url https://flashinfer.ai/whl/cu130

Download the plugins:

bash

hf download elyza/ELYZA-Thinking-1.0-llm-jp-4-33b --include "vllm_plugins/*" --local-dir .

hf download elyza/ELYZA-Thinking-1.0-llm-jp-4-33b --include "vllm_plugins/*" --local-dir .

Start the server:

bash

vllm serve elyza/ELYZA-Thinking-1.0-llm-jp-4-33b \  --max-model-len 65536 \  --trust-remote-code \  --reasoning-parser-plugin vllm_plugins/multi_parser_loader.py \  --reasoning-parser llmjp4 \  --tool-parser-plugin vllm_plugins/llmjp_harmony_tool_parser.py \  --tool-call-parser llmjp_harmony \  --enable-auto-tool-choice

vllm serve elyza/ELYZA-Thinking-1.0-llm-jp-4-33b \  --max-model-len 65536 \  --trust-remote-code \  --reasoning-parser-plugin vllm_plugins/multi_parser_loader.py \  --reasoning-parser llmjp4 \  --tool-parser-plugin vllm_plugins/llmjp_harmony_tool_parser.py \  --tool-call-parser llmjp_harmony \  --enable-auto-tool-choice

Send a request. The answer is returned in content, and the reasoning separately in the message's reasoning field:

bash

curl http://localhost:8000/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b",    "messages": [{"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}],    "temperature": 0.6,    "top_p": 0.95  }'

curl http://localhost:8000/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b",    "messages": [{"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}],    "temperature": 0.6,    "top_p": 0.95  }'

bash

curl http://localhost:8000/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b",    "messages": [{"role": "user", "content": "東京の天気を教えて"}],    "tools": [{      "type": "function",      "function": {        "name": "get_weather",        "description": "Get the current weather of a city",        "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}      }    }]  }'

curl http://localhost:8000/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b",    "messages": [{"role": "user", "content": "東京の天気を教えて"}],    "tools": [{      "type": "function",      "function": {        "name": "get_weather",        "description": "Get the current weather of a city",        "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}      }    }]  }'

Transformers

Works with the latest Transformers (tested with 4.57.6 and 5.18.0):

bash

pip install -U transformers accelerate torch

python

import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM model_name = "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b" tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(    model_name,    device_map="auto",    dtype=torch.bfloat16,    trust_remote_code=True,)model.eval() messages = [{"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}]inputs = tokenizer.apply_chat_template(    messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device) with torch.no_grad():    output_ids = model.generate(        **inputs,        max_new_tokens=16384,  # thinking model: leave room for the reasoning before the answer        do_sample=True,        temperature=0.6,        top_p=0.95,    )[0][inputs["input_ids"].shape[-1]:] for m in tokenizer.parse_harmony_message(output_ids.tolist()):    print(f"[{tokenizer.decode(m.channel.token_ids)}]", tokenizer.decode(m.content.token_ids))

import torchfrom transformers import AutoTokenizer, AutoModelForCausalLM model_name = "elyza/ELYZA-Thinking-1.0-llm-jp-4-33b" tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)model = AutoModelForCausalLM.from_pretrained(    model_name,    device_map="auto",    dtype=torch.bfloat16,    trust_remote_code=True,)model.eval() messages = [{"role": "user", "content": "仕事の熱意を取り戻すためのアイデアを5つ挙げてください。"}]inputs = tokenizer.apply_chat_template(    messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device) with torch.no_grad():    output_ids = model.generate(        **inputs,        max_new_tokens=16384,  # thinking model: leave room for the reasoning before the answer        do_sample=True,        temperature=0.6,        top_p=0.95,    )[0][inputs["input_ids"].shape[-1]:] for m in tokenizer.parse_harmony_message(output_ids.tolist()):    print(f"[{tokenizer.decode(m.channel.token_ids)}]", tokenizer.decode(m.content.token_ids))

💡 Tip Recommended sampling: temperature=0.6, top_p=0.95 (default in generation_config.json).

Acknowledgements

How to Cite

text

@misc{elyza2026thinking,    title={elyza/ELYZA-Thinking-1.0-llm-jp-4-33b},    url={https://huggingface.co/elyza/ELYZA-Thinking-1.0-llm-jp-4-33b},    author={Xin Zhao, Congda Ma, Satoshi Tohda, Tasavat Trisitichoke, Akira Sasaki, Masato Hirakawa, Kota Kakiuchi},    year={2026},}

@misc{elyza2026thinking,    title={elyza/ELYZA-Thinking-1.0-llm-jp-4-33b},    url={https://huggingface.co/elyza/ELYZA-Thinking-1.0-llm-jp-4-33b},    author={Xin Zhao, Congda Ma, Satoshi Tohda, Tasavat Trisitichoke, Akira Sasaki, Masato Hirakawa, Kota Kakiuchi},    year={2026},}

Citations

text

@misc{llm-jp-4-33b-base,    title = {llm-jp/llm-jp-4-33b-base},    url = {https://huggingface.co/llm-jp/llm-jp-4-33b-base},    author = {Research and Development Center for Large Language Models},    year = {2026}}

@misc{llm-jp-4-33b-base,    title = {llm-jp/llm-jp-4-33b-base},    url = {https://huggingface.co/llm-jp/llm-jp-4-33b-base},    author = {Research and Development Center for Large Language Models},    year = {2026}}