Jais-3-0.9B

Jais-3-0.9B is the compact dense member of the Jais-3 family: a 0.9B-class decoder-only model with a 128K context window.

Other formats are available in the IFM/Jais-3 collection:

• GGUF for llama.cpp and Ollama: IFM/Jais-3-0.9B-GGUF
• MLX for Apple silicon:
  • IFM/Jais-3-0.9B-MLX-4bit
  • IFM/Jais-3-0.9B-MLX-8bit
  • IFM/Jais-3-0.9B-MLX-bf16

Benchmark Results

		Reference models
	Jais-3-0.9B	Qwen3.5-2B	Falcon-H1-1.5B-Deep-Instruct	Hala 1.2B	tiny-aya-global
# Params	0.9B	2B	1.5B	1.2B	3B
# Activated params	0.9B	2B	1.5B	1.2B	3B
Architecture	Dense	Dense	Dense	Dense	Dense
Open-ended Generation
AraGen
AraGen v3 (3C3H) 09-2025	32.7	15.3	12.2	25.0	21.4
Correctness	39.4	20.0	15.5	28.2	27.3
Completeness	38.6	19.3	15.0	27.1	25.2
Conciseness	10.9	2.6	3.1	14.4	6.5
Helpfulness	33.2	15.4	11.4	25.8	21.1
Honesty	35.8	15.4	13.1	26.8	22.1
Harmlessness	38.7	19.0	15.4	27.5	26.5
Knowledge & Reasoning
General Knowledge & Comprehension
Arabic MMLU* Accuracy	70.4	63.1	48.7	45.0	55.9
Belebele Accuracy	81.6	67.6	57.8	39.9	62.2
STEM & Math
AraSTEM* Accuracy	81.6	83.9	62.5	48.3	58.5
Arabic GSM8K* Accuracy	80.9	75.6	65.4	57.3	59.6
GATmath* Accuracy	87.4	87.4	40.2	29.6	31.4
Professional Domains
MedAraBench* Accuracy	48.1	54.3	32.3	32.3	39.5
Arabic Legal (HELM)* Accuracy	58.7	49.1	48.8	41.1	52.2
Arabic Language & Culture
Islamic Knowledge & Arab Culture
IslamicQA Accuracy (mean of the 4 below)	66.2	43.0	48.2	35.9	44.2
PaLM-X 2025-2* Accuracy	78.1	62.0	62.1	50.0	71.6
QIAS-1* Accuracy	61.1	26.1	32.1	16.2	23.1
QIAS-2* Accuracy	65.0	51.5	56.0	41.9	54.7
IslamicEval2025-1B* Accuracy	60.6	32.5	42.4	35.5	27.3
Arab Culture Accuracy (mean of the 3 below)	73.3	54.8	54.4	46.9	61.5
ArabCulture* Accuracy	74.0	63.8	61.7	60.6	68.6
AraDiCE-Culture* Accuracy	72.6	46.9	49.2	30.7	55.3
Arabic Cuisine^ Accuracy	73.4	53.6	52.4	49.5	60.6
Dialects
Dialectal Arabic MMLU* Accuracy	63.7	59.5	42.3	40.3	47.7
Dialect Identification (QADI) Accuracy	65.9	24.2	23.4	23.0	38.8
Alyah* Accuracy	73.8	53.2	54.8	53.2	68.1
Linguistics & Literature
Arabic MMLU with diacritics Accuracy	65.1	49.4	34.7	31.3	45.6
FannOrFlop-MCQ Accuracy	59.3	43.6	38.5	31.3	48.5
Arabic Linguistics^ Accuracy (mean of the 2 below)	85.5	54.7	33.3	32.6	38.2
Diacritization Accuracy	83.6	65.1	26.0	32.9	36.7
Parsing Accuracy	87.4	44.4	40.6	32.2	39.6
Assistant Capabilities
Instruction Following
Arabic IFBench^ Accuracy (prompt-level strict)	91.1	33.1	16.2	14.6	21.5
Arabic IFEval Accuracy	71.3	62.6	49.3	44.1	58.2
Factuality & Hallucination
HalluScore Accuracy	28.0	25.8	6.0	10.1	20.1
SimpleQA Verified (Ar) Aggregate (2/3 F-score + 1/3 non-hallucination rate)	27.1	7.9	2.6	3.1	3.1
F-score	1.0	2.7	2.0	3.1	3.1
Non-hallucination rate	79.3	18.2	3.8	3.2	3.2

Starred (*) benchmarks have been audited for sample quality. Items with ambiguous questions, invalid answer options, or indefensible gold labels were excluded, since they can penalize a correct answer or reward a lucky guess. Scores on these benchmarks are reported on the cleaned sets. The audit methodology, the excluded samples and the verified benchmarks are available in the IFM/Arabic-Benchmark-Auditing collection.

Benchmarks marked with ^ will be released by IFM soon.

Quickstart

Serving

vLLM (vllm>=0.30.0) · vLLM recipe:

vllm serve inception42/Jais-3-0.9B \
  --trust-remote-code \
  --dtype bfloat16 \
  --max-model-len 131072 \
  --hf-overrides '{"rope_parameters":{"rope_type": "yarn", "factor": 16, "original_max_position_embeddings": 8192, "rope_theta": 1000000, "beta_fast": 128, "beta_slow": 4}}' \
  --gpu-memory-utilization 0.85 \
  --tensor-parallel-size 1 \
  --reasoning-parser k2_horizon \
  --enable-auto-tool-choice \
  --tool-call-parser k2_horizon

Use an exact branch name from the inventory with vLLM's --revision option.

SGLang (sglang>=0.5.20) · SGLang recipe:

sglang serve \
  --model-path inception42/Jais-3-0.9B \
  --tp 1 \
  --dtype bfloat16 \
  --attention-backend fa3 \
  --reasoning-parser k2_horizon \
  --host 0.0.0.0 \
  --port 30000

API Usage

Recommended settings: reasoning_effort="high", the sampling hyperparameters in the table below, and at least 32,768 output tokens. Reasoning depth is selected per request through chat_template_kwargs. Thinking is returned in reasoning_content and the answer in content.

Parameter	Default	Thinking mode, English	Thinking mode, Arabic
temperature	0.2	1.0	0.7
top_p	0.95	0.95	0.95
top_k	20	20	20

from openai import OpenAI
client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY")
response = client.chat.completions.create(
    model="inception42/Jais-3-0.9B",
    messages=[{"role": "user", "content": "لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل."}],
    temperature=0.7,
    top_p=0.95,
    max_tokens=32768,
    extra_body={"top_k": 20, "chat_template_kwargs": {"reasoning_effort": "high"}},
)
message = response.choices[0].message
print("Reasoning:", getattr(message, "reasoning_content", None))
print("Answer:", message.content)

Transformers

Validated with Transformers 5.15.0, PyTorch 2.13.0, Safetensors 0.8.0.

from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "inception42/Jais-3-0.9B"
tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_id, device_map="auto", dtype="bfloat16", low_cpu_mem_usage=True, trust_remote_code=True
)
inputs = tokenizer("لدينا عددان مجموعهما 50 وحاصل ضربهما 600. ما هما العددان؟ اشرح خطوات الحل.", return_tensors="pt").to(model.device)
inputs.pop("token_type_ids", None)
outputs = model.generate(**inputs, max_new_tokens=32768, temperature=0.7, top_p=0.95, top_k=20, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Release Artifacts

The table below lists the release artifacts for Jais-3-0.9B.

Artifact Index

Artifact	Link
Model card	Hugging Face
Training logs	W&B
Blog post	Jais 3: Introducing Arabic-centric Agentic AI
Checkpoints	Checkpoint inventory
Technical report	Coming soon
Code repository	Coming soon

Checkpoint Inventory

Model repository: inception42/Jais-3-0.9B

Branch names below refer to this repository.

Checkpoint	Branch / repository
CPT Checkpoint	CPT
Cooldown Checkpoint	Cooldown
FTPO Checkpoint	FTPO
SFT Checkpoint	SFT

Best Practices

1. Reasoning effort: always high. All reported results use high reasoning effort. Pass {"chat_template_kwargs": {"reasoning_effort": "high"}} on every request. No reasoning ({"reasoning_effort": "low"}) is not recommended except for research on reasoning efforts.
2. Sampling parameters. Use the values in the table above: temperature=0.2, top_p=0.95, top_k=20 by default. In thinking mode, set temperature=1.0 for English and temperature=0.7 for Arabic. Keep min_p=0.0, presence_penalty=0.0 and repetition_penalty=1.0.
3. Serving. Use the validated SGLang recipe above: BF16, TP=1 on a single H100 node, FlashAttention-3.
4. Parsers. Enable the k2_horizon reasoning parser for chat, and add the k2_horizon tool-call parser for agent use. Leave both off for plain completion-style generation.

Intended Use and Limitations

Intended use: The model is designed for Arabic and English assistant and agentic applications, including reasoning and tool-use tasks, as well as for research and experimentation.

Out of scope: The model should not be used in ways that violate applicable laws or regulations. The following uses are also considered out of scope:

• Harmful or malicious use: generating hateful, extremist, discriminatory, or deliberately misleading content, or using the model to support illegal or harmful activities.
• Sensitive information: using the model to generate, expose, or infer personal, confidential, or otherwise sensitive information about individuals or organizations.
• Unsupported languages: applications that depend on reliable performance in languages other than Arabic or English.
• High-stakes decisions: relying on the model to make medical, legal, financial, or safety-critical decisions without appropriate human oversight.

Limitations: The model can produce inaccurate or made-up content and may reflect biases present in its training data. Evaluate it for your use case and add safeguards before deploying it.

Citation

@misc{jais3_2026,
  title  = {Jais 3: Introducing Arabic-centric Agentic AI},
  author = {{IFM Team}},
  year   = {2026},
  url    = {https://ifm.ai/blog/jais3/},
}