Hunyuan-A13B-Instruct

tencent tencent-hunyuan-a13b out 16 months ago

80.4B parameters, 13B active, text in, text out, 32K tokens of context. 1 provider serves it from $0.14 per million input tokens.

Verified 2026-10-25 Rank #481 in LLMs · score 77 method v1 Sources 4 Downloads 30d 69k · #359 of 619 tracked

Can I run it?48GB at INT4Estimatefits a 64 GB Mac (unified memory)method

What does it cost?$0.14/1M incheapest of 1SiliconFlowfp82026-10-24

Can I use it?tencent-hunyuan-a13bCustom license; read the repo LICENSE before commercial hostinglicence file2026-10-24

Do people use it?69kdownloads, 30 days#359 of 619 we track · below the median+0 likes, downloads −4% this week795 likesHugging Face2026-10-24

Worth reading

Picked by rule from 15 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Most discussed · 148 HN pts

Tencent Hunyuan-Large

Tencent releases Hunyuan-Large, the largest open-source Transformer-based MoE model with 389 billion total parameters and 52 billion active parameters.

The TRT-LLM solution is widely used in Tencent's Hunyuan project, with plans to release it in the near future.

github.com 2024-11-05 148 HN pts · 103 comments 4k words about the Hunyuan family family

Latest · 15 months ago

Tencent Releases Hunyuan World Model

Tencent released HunyuanWorld-1.0, an open-source model for generating immersive 3D worlds from text or images.

Engineers must install the draco library to export generated scenes in draco format.
Inference speed and memory usage can be optimized by enabling fp8_gemm and fp8_attention flags.

github.com 2025-07-28 44 HN pts · 6 comments 1.5k words about the Hunyuan family family

15 best of 15 tracked

The 2 picks above, plus these 13. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

ArticleTencent's 'Hunyuan-T1'–The First Mamba-Powered Ultra-Large Modelllm.hunyuan.tencent.com · 2025-03-21 · 298 HN pts · official
PaperHunyuan-Large: An Open-Source Moe Model with 52B Activated Parametersarxiv.org · 2024-11-05 · 7 HN pts
ArticleHunyuan T1 Mamba Reasoning model beats R1 on speed and metricstencent.github.io · 2025-03-22 · 26 HN pts
ArticleTencent Hunyuan open-sources its first hybrid AI modeltechinasia.com · 2025-06-27 · 8 HN pts
ArticleTencent Hunyuan video: open-source text-to-video generation modelaivideo.hunyuan.tencent.com · 2024-12-03 · 5 HN pts · official
CodeHunyuan-Large pre-trained model achieves the best overall performancegithub.com · 2024-11-05 · 3 HN pts
CodeTencent Hunyuan-A13Bgithub.com · 2025-06-28 · 7 HN pts
CodeHunyuan-DiT: Multi-Resolution Diffusion Transformer with Chinese Understandinggithub.com · 2024-05-17 · 5 HN pts
CodeTencent launches a new generation of fast thinking model Hunyuan Turbo Sgithub.com · 2025-02-27
CodeTencent Hunyuan-Largegithub.com · 2025-08-18
DiscussionI will train and open-source 50 SFW Hunyuan Video LoRAs. Request anythingnews.ycombinator.com · 2025-02-20
ArticleTencent HunYuan 3D Generative Model v2.5, Significantly Enhanced Precisionaibase.com · 2025-04-29
ArticleIn-Depth Comparison of Open Video Models: Hunyuan, LTX, CogVideoXdrsandor.net · 2024-12-22

Show all 13

Where to run it

Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.

#	Cheapest live row	In /1M	Out /1M	Context	Quant	Seen
1	SiliconFlow	$0.14	$0.57	128K	fp8	Oct 24

All 1 provider →

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

Served by 1 API provider; 17 quantized builds for local use.

As a service1SiliconFlow
On own hardware17quantized builds
Built on129 fine-tunes · 3 adapters · 0 merges
In apps5Hugging Face Spaces

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 64 GB Mac (unified memory) at INT4

About 48 GB at INT4, 97 GB at FP8, 193 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

Most coverage came in the first two weeks

Teal is the weights (2025-06-25), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.

What people built on it

29 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

29 derivatives on the Hub, most of them quantized

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

GGUF unsloth/Hunyuan-A13B-Instruct-GGUF for llama.cpp, Ollama, LM Studio 49.6k downloads MLX mlx-community/Hunyuan-A13B-Instruct-4bit for Apple silicon 2.2k downloads GPTQ tencent/Hunyuan-A13B-Instruct-GPTQ-Int4 for vLLM, SGLang 179.5k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

22 rows

Benchmarks 22 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (22)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.

Benchmark	Value	Setting / notes	Reported by	Date
author-reported AIME 2024 · Reason through hard problems	87.3	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported AIME 2025 · Reason through hard problems	76.8	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported ArtifactsBench	43	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported BBH · Reason through hard problems	87.56	Hunyuan-A13B	model author	2026-10-07
author-reported CRUX-I	70.13	Hunyuan-A13B	model author	2026-10-07
author-reported CRUX-O	77	Hunyuan-A13B	model author	2026-10-07
author-reported Fullstackbench	67.8	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported GPQA Diamond · Reason through hard problems	49.12	Hunyuan-A13B	model author	2026-10-07
author-reported GPQA Diamond · Reason through hard problems	71.2	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported GSM8K · Reason through hard problems	91.83	Hunyuan-A13B	model author	2026-10-07
author-reported HumanEval · Write code with agents	78.64	Hunyuan-A13B	model author	2026-10-07
author-reported LiveCodeBench · Write code with agents	63.9	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported MATH-500 · Reason through hard problems	72.35	Hunyuan-A13B	model author	2026-10-07
author-reported MATH-500 · Reason through hard problems	91.17	Hunyuan-A13B	model author	2026-10-07
author-reported MATH-500 · Reason through hard problems	94.3	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported MBPP · Write code with agents	83.86	Hunyuan-A13B	model author	2026-10-07
author-reported MMLU · Reason through hard problems	88.17	Hunyuan-A13B	model author	2026-10-07
author-reported MMLU · Reason through hard problems	87.67	Hunyuan-A13B	model author	2026-10-07
author-reported MMLU-Pro · Reason through hard problems	67.23	Hunyuan-A13B	model author	2026-10-07
author-reported MultiPL-E · Write code with agents	69.33	Hunyuan-A13B	model author	2026-10-07
author-reported OlympiadBench	82.7	Hunyuan-A13B-Instruct	model author	2026-10-07
author-reported SuperGPQA · Reason through hard problems	41.32	Hunyuan-A13B	model author	2026-10-07

Use-case evidence

5 use cases

What it is good at 5 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

21 fields

Listed in LLMs · MoE

Facts

Hunyuan-A13B-Instruct has 80.4B parameters (13B active per token). (safetensors metadata, observed 2026-10-24)
Hunyuan-A13B-Instruct is a mixture-of-experts model with 64 experts. (config.json, observed 2026-10-24)
Context length is 32,768 tokens (32K). (config.json max_position_embeddings, observed 2026-10-24)
License: tencent-hunyuan-a13b (friction: review). Custom license; read the repo LICENSE before commercial hosting. (license file, observed 2026-10-24)
Created 2025-06-25, last modified 2025-08-21. (HF API, observed 2026-10-24)

Pipeline tag	text-generation
Library	transformers
Architectures	HunYuanMoEV1ForCausalLM
Model type	hunyuan_v1_moe
Dtype	bfloat16
Params by dtype	BF16: 80,384,794,624 · F32: 8,388,608
Size band	XL
MoE	yes
Experts	64
Context length	32,768 (config.json max_position_embeddings)
License	tencent-hunyuan-a13b (other)
Hub snapshot	2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · HunYuanMoEV1ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for Hunyuan-A13B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)	193 GB · Multi-GPU node (<640 GB)
FP8 (1 byte/param)	97 GB · Multi-GPU node (<640 GB)
INT4 (0.5 bytes/param)	48 GB · Single high-end GPU (<80 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.

All variants

29 variants

Lineage

Original model Family: tencent/Hunyuan 29 derivatives on the Hub

quantized: 17finetune: 9adapter: 3

Top variants (29 shown, ranked by downloads)

Variant	Relation	Quant	Params	Band	License	Author	Likes	Downloads
tencent/Hunyuan-A13B-Instruct-GPTQ-Int4	quantized	GPTQ	80.4B	XL	other	tencent	53	179.5k
unsloth/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	unsloth	48	49.6k
bullerwins/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	bullerwins	24	22.4k
bartowski/tencent_Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	bartowski	5	17.5k
lmstudio-community/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	lmstudio-community	1	9.2k
gabriellarson/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	gabriellarson	2	5.6k
mradermacher/Hunyuan-A13B-Instruct-i1-GGUF	quantized	GGUF	—	—	other	mradermacher	0	4.2k
DevQuasar/tencent.Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	—	DevQuasar	1	3.9k
ubergarm/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	ubergarm	13	2.7k
mlx-community/Hunyuan-A13B-Instruct-4bit	quantized	-4bit	80.4B	XL	other	mlx-community	4	2.2k
mradermacher/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	other	mradermacher	0	1.5k
mlx-community/Hunyuan-A13B-Instruct-5bit	quantized	—	80.4B	XL	other	mlx-community	3	1.2k
yujiepan/hunyuan-moe-tiny-random	finetune	—	8.67M	S	—	yujiepan	0	937
Face314/Hunyuan-A13B-Instruct-MXFP4_MOE	quantized	GGUF	—	—	other	Face314	0	721
mlx-community/Hunyuan-A13B-Instruct-4bit-DWQ	quantized	-4bit	80.4B	XL	other	mlx-community	0	706
unsloth/Hunyuan-A13B-Instruct	finetune	—	80.4B	XL	other	unsloth	2	480
bullerwins/Hunyuan-A13B-Instruct-hf	finetune	—	80.4B	XL	other	bullerwins	6	242
mlx-community/Hunyuan-A13B-Instruct-8bit	quantized	-8bit	80.4B	XL	other	mlx-community	0	439
mlx-community/Hunyuan-A13B-Instruct-mixed-6-8bit	quantized	-8bit	80.4B	XL	other	mlx-community	0	413
mlx-community/Hunyuan-A13B-Instruct-bf16	finetune	-bf16	80.4B	XL	other	mlx-community	0	353
tiny-random/hunyuan-moe	finetune	—	8.67M	S	—	tiny-random	1	201
redponike/Hunyuan-A13B-Instruct-GGUF	quantized	GGUF	—	—	—	redponike	0	213
nqzfaizal77ai/heaven-hyperion-reinit-542m-zero	finetune	—	542M	S	—	nqzfaizal77ai	0	101
francescortu/DistillDetect-ctrl-hunyuan-a13b-from-qwen35-lora	adapter	—	—	—	—	francescortu	0	32
francescortu/DistillDetect-ctrl-hunyuan-a13b-from-gptoss-lora	adapter	—	—	—	—	francescortu	0	29
francescortu/DistillDetect-ctrl-hunyuan-a13b-from-gemma-lora	adapter	—	—	—	—	francescortu	0	27
Shamoddfh/cb	finetune	—	—	—	—	Shamoddfh	0	0
gusjohnalkoding/pirate_chatbot	finetune	—	—	—	llama3.3	gusjohnalkoding	0	0
MgrEvil/Julia	finetune	—	—	—	artistic-2.0	MgrEvil	0	0

Repository changelog

30 commits

Repository changelog 30 commits

Date	Commit	SHA
2025-08-21	Update config.json	290ddb9
2025-08-21	Delete tokenization_hy.py	9f1d108
2025-08-21	Delete hy.tiktoken	9aa9e81
2025-08-21	Delete pytorch_model.bin.index.json	5de8f3a
2025-08-21	Upload tokenizer_config.json with huggingface_hub	600f638
2025-08-21	Upload tokenizer.json with huggingface_hub	052ecbd
2025-08-21	Upload pytorch_model.bin.index.json with huggingface_hub	cbe7cd3
2025-08-21	Upload generation_config.json with huggingface_hub	c3a2ea2
2025-08-21	Upload config.json with huggingface_hub	1ad3653
2025-07-22	Update config.json	f21d329
2025-07-08	Update README_CN.md	7302638
2025-07-08	Update README.md	2c12fe4
2025-07-02	update-readme (#30)	798e8a3
2025-07-01	Delete pytorch_model.bin.index.json	65f74cc
2025-07-01	Delete pytorch_model-00004-of-00004.bin	d1856a0
2025-07-01	Delete pytorch_model-00003-of-00004.bin	8d0586c
2025-07-01	Delete pytorch_model-00002-of-00004.bin	df8f53e
2025-07-01	Delete pytorch_model-00001-of-00004.bin	5b68752
2025-07-01	Upload model-00020-of-00033.safetensors with huggingface_hub	c629f12
2025-07-01	Upload model-00013-of-00033.safetensors with huggingface_hub	1c038aa
2025-07-01	Upload model-00015-of-00033.safetensors with huggingface_hub	b5507dd
2025-07-01	Upload model-00017-of-00033.safetensors with huggingface_hub	7a5074c
2025-07-01	Upload model-00012-of-00033.safetensors with huggingface_hub	83c6eda
2025-07-01	Upload model-00026-of-00033.safetensors with huggingface_hub	a27f874
2025-07-01	Upload model-00008-of-00033.safetensors with huggingface_hub	e170033
2025-07-01	Upload model-00027-of-00033.safetensors with huggingface_hub	00013f9
2025-07-01	Upload model-00025-of-00033.safetensors with huggingface_hub	b6789f2
2025-07-01	Upload model-00011-of-00033.safetensors with huggingface_hub	ef38713
2025-07-01	Upload model-00010-of-00033.safetensors with huggingface_hub	b9bf2d2
2025-07-01	Upload model-00024-of-00033.safetensors with huggingface_hub	9a3115a

All providers

1 provider · 1 endpoint

Where to run 1 providers · 1 endpoints

Hosted endpoints observed for Hunyuan-A13B-Instruct, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.14/M.

Provider	Model ref	Context	Quant	$/M in	$/M out	tok/s	TTFT	Tools	Status	Observed
SiliconFlow openrouter	tencent/hunyuan-a13b-instruct	128K	fp8	$0.14	$0.57	—	—	—	live	2026-10-24

Articles and coverage

15 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official2

Tencent's 'Hunyuan-T1'–The First Mamba-Powered Ultra-Large Model

llm.hunyuan.tencent.com2025-03-21officialHN front pageabout the Hunyuan familyHN 298 pts · 188 comments37 words

Tencent Hunyuan video: open-source text-to-video generation model

aivideo.hunyuan.tencent.com2024-12-03officialabout the Hunyuan familyHN 5 pts · 0 comments690 words

Analysis8

Tencent Hunyuan-Large

github.com2024-11-05authorityHN front pageabout the Hunyuan familyHN 148 pts · 103 comments4,019 words

Tencent releases Hunyuan-Large, the largest open-source Transformer-based MoE model with 389 billion total parameters and 52 billion active parameters. The model utilizes Grouped Query Attention (GQA) and Cross-Layer Attention (CLA) strategies to achieve a 50% savings in the KV-Cache portion and supports text sequences up to 256K. While the TRT-LLM solution surpasses vLLM by over 30%, full fine-tuning requires at least 32 GPUs, and LoRA fine-tuning requires at least 8 GPUs.written, reviewed, cited

The TRT-LLM solution is widely used in Tencent's Hunyuan project, with plans to release it in the near future.

Tencent Releases Hunyuan World Model

github.com2025-07-28authorityabout the Hunyuan familyHN 44 pts · 6 comments1,474 words

Tencent released HunyuanWorld-1.0, an open-source model for generating immersive 3D worlds from text or images. The framework uses a semantically layered 3D mesh representation and supports running on Consumer-grade GPUs such as 4090. Engineers can deploy the 478MB checkpoints using Python 3.10 and PyTorch 2.5.0+cu124.written, reviewed, cited

Engineers must install the draco library to export generated scenes in draco format.
Inference speed and memory usage can be optimized by enabling fp8_gemm and fp8_attention flags.

Hunyuan-Large: An Open-Source Moe Model with 52B Activated Parameters

arxiv.org2024-11-05authorityabout the Hunyuan familyHN 7 pts · 1 comments391 words

Tencent released Hunyuan-Large, the largest open-source Transformer-based mixture of experts model, featuring 389 billion total parameters and 52 billion activation parameters. It handles up to 256K tokens and outperforms Llama3.1-70B while matching Llama3.1-405B performance across benchmarks like coding and reasoning. Key practices include large-scale synthetic data, a mixed expert routing strategy, a key-value cache compression technique, and an expert-specific learning rate strategy.written, reviewed, cited

Hunyuan T1 Mamba Reasoning model beats R1 on speed and metrics

tencent.github.io2025-03-22about the Hunyuan familyHN 26 pts · 2 comments775 words

Tencent released Hunyuan-T1, an official reasoning model based on the TurboS base architecture that utilizes reinforcement learning to achieve performance comparable to OpenAI's O1 on MMLU-PRO with a score of 87.2. The model invested 96.7% of its computing power in reinforcement learning training and achieves a decoding speed 2 times faster under the same deployment conditions. It scored 69.3 on GPQA-diamond and 91.9 in the ArenaHard task, demonstrating strong adaptability in instruction-following and tool utilization.written, reviewed, cited

The model is based on the world's first ultra-large-scale Hybrid-Transformer-Mamba MoE architecture.
Training stability improved by over 50% using classic reinforcement learning strategies like data replay.
The reward system combines self-rewarding based on an early version of T1-preview with a reward model.

Tencent Hunyuan open-sources its first hybrid AI model

techinasia.com2025-06-27about the Hunyuan familyHN 8 pts · 4 comments

Hunyuan-Large pre-trained model achieves the best overall performance

github.com2024-11-05authorityabout the Hunyuan familyHN 3 pts · 2 comments1,623 words

Tencent Hunyuan-A13B

github.com2025-06-28authorityHN 7 pts · 0 comments1,748 words

Hunyuan-DiT: Multi-Resolution Diffusion Transformer with Chinese Understanding

github.com2024-05-17authorityabout the Hunyuan familyHN 5 pts · 0 comments4,398 words

All coverage tracked for this model (15) — append-only, nothing is ever removed

Title	Source	Published	First seen	Signals
Tencent Hunyuan-Large	github.com	2025-08-18	2026-10-07	hn
Tencent Releases Hunyuan World Model	github.com	2025-07-28	2026-10-07	hn
Tencent Hunyuan-A13B	github.com	2025-06-28	2026-10-07	hn
Tencent Hunyuan open-sources its first hybrid AI model	techinasia.com	2025-06-27	2026-10-07	hn
Tencent HunYuan 3D Generative Model v2.5, Significantly Enhanced Precision	aibase.com	2025-04-29	2026-10-07	hn
Hunyuan T1 Mamba Reasoning model beats R1 on speed and metrics	tencent.github.io	2025-03-22	2026-10-07	hn
Tencent's 'Hunyuan-T1'–The First Mamba-Powered Ultra-Large Model	llm.hunyuan.tencent.com official	2025-03-21	2026-10-07	hn
Tencent launches a new generation of fast thinking model Hunyuan Turbo S	github.com	2025-02-27	2026-10-07	hn
I will train and open-source 50 SFW Hunyuan Video LoRAs. Request anything	news.ycombinator.com	2025-02-20	2026-10-07	hn
In-Depth Comparison of Open Video Models: Hunyuan, LTX, CogVideoX	drsandor.net	2024-12-22	2026-10-07	hn
Tencent Hunyuan video: open-source text-to-video generation model	aivideo.hunyuan.tencent.com official	2024-12-03	2026-10-07	hn
Tencent Hunyuan-Large	github.com	2024-11-05	2026-10-07	hn
Hunyuan-Large: An Open-Source Moe Model with 52B Activated Parameters	arxiv.org	2024-11-05	2026-10-07	hn
Hunyuan-Large pre-trained model achieves the best overall performance	github.com	2024-11-05	2026-10-07	hn
Hunyuan-DiT: Multi-Resolution Diffusion Transformer with Chinese Understanding	github.com	2024-05-17	2026-10-07	hn

Similar models

3 models

Similar models

Qwen3-Next-80B-A3B-Instruct · 81.3B
same size band (XL), also MoE, same task (text-generation), more providers, permissive license
Qwen3-Coder-Next · 79.7B
same size band (XL), also MoE, same task (text-generation), more providers, permissive license
Qwen3-Next-80B-A3B-Thinking · 81.3B
same size band (XL), also MoE, same task (text-generation), permissive license

Sources

4 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:39 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:10 UTC · Last verified 2026-10-25

Follow Hunyuan-A13B-Instruct

Watch this model

/feeds/models/tencent--Hunyuan-A13B-Instruct.xml(RSS 2.0) · every change to Hunyuan-A13B-Instruct
/feeds/families/tencent--Hunyuan.xml(RSS 2.0) · every change across the Hunyuan family

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.