# OpenAI Unveils Chestnut and Hazelnut AI Models

> Published 2026-04-01, updated 2026-09-05 · https://www.promptzone.com/rohan_moreau/openai-unveils-chestnut-and-hazelnut-ai-models-32n2

OpenAI has dropped two new AI models, **Chestnut** and **Hazelnut**, targeting distinct use cases in the generative AI space. Announced recently, these models aim to push boundaries in text generation and multimodal capabilities with competitive pricing and performance metrics. Let’s break down what each brings to the table for developers and researchers.

> **Model:** Chestnut | **Parameters:** 13B | **Speed:** 45 tokens/sec 
> **Price:** $0.05 per 1M tokens | **Available:** OpenAI API | **License:** Commercial

> **Model:** Hazelnut | **Parameters:** 7B | **Speed:** 60 tokens/sec 
> **Price:** $0.02 per 1M tokens | **Available:** OpenAI API | **License:** Commercial

## Chestnut: Power for Complex Tasks
**Chestnut**, with its **13B** parameters, is built for heavy lifting in natural language processing. It clocks in at **45 tokens per second**, making it a solid choice for applications requiring deep contextual understanding, such as long-form content creation or intricate dialogue systems. Early testers report that Chestnut excels in maintaining coherence over extended text outputs, a common challenge for smaller models.

At **$0.05 per 1M tokens**, it’s priced for enterprise users who need robust performance without breaking the bank. The model is accessible via the [OpenAI API](https://openai.com/api/), ensuring seamless integration into existing workflows.

> **Bottom line:** Chestnut offers a balance of power and affordability for demanding NLP tasks.


![OpenAI Unveils Chestnut and Hazelnut AI Models](https://promptzone-community.s3.amazonaws.com/uploads/articles/z9ccdgr61mggpkp88t19.png)

## Hazelnut: Speed and Efficiency
On the other end, **Hazelnut** targets lightweight, high-speed applications with **7B** parameters and a blazing **60 tokens per second**. This model is ideal for real-time use cases like chatbots or quick content drafting where latency is critical. Users note its responsiveness, especially in mobile or edge deployments with limited compute resources.

Priced at just **$0.02 per 1M tokens**, Hazelnut undercuts many competitors in the budget segment. Like Chestnut, it’s available through the [OpenAI API](https://openai.com/api/), offering flexibility for developers scaling smaller projects.

## Head-to-Head Comparison
| Feature | Chestnut | Hazelnut |
|------------------|----------------|----------------|
| Parameters | 13B | 7B |
| Speed | 45 tokens/sec | 60 tokens/sec |
| Price per 1M tokens | $0.05 | $0.02 |
| Best Use Case | Complex NLP | Real-time apps |

This table highlights the trade-offs: Chestnut for depth, Hazelnut for speed. Developers choosing between them should weigh project requirements against budget and latency constraints.

## Technical Deep Dive
{% details "VRAM and Deployment Notes" %}
- Chestnut requires approximately **26GB VRAM** for full precision, though quantization can drop this to **16GB** on consumer-grade GPUs.
- Hazelnut is lighter, needing **14GB VRAM** unquantized and as low as **10GB** with optimization.
- Both models support fine-tuning via OpenAI’s platform, though specific compute costs for training runs are not yet public.
{% enddetails %}

## Community Buzz and Use Cases
Feedback from early adopters suggests both models are finding niches fast. Chestnut is gaining traction among developers building legal or academic writing tools, thanks to its knack for nuanced language. Hazelnut, meanwhile, is popping up in customer service bots, where its **60 tokens/sec** speed keeps interactions snappy. Some users have flagged Chestnut’s higher VRAM demands as a barrier for smaller setups, but quantization options are easing the pain.

> **Bottom line:** Hazelnut’s low cost and speed make it a go-to for lightweight apps, while Chestnut targets power users.

## What’s Next for OpenAI’s Lineup
With Chestnut and Hazelnut, OpenAI is clearly segmenting its offerings to capture both high-end and budget-conscious markets. As competition heats up in the AI space, these models could set a new benchmark for balancing cost and capability. Keep an eye on how the community adapts these tools for specialized applications in the coming months.

<!-- pz-related-guides -->
## Related guides on PromptZone

- [Local LLMs 2026: Run Llama, Mistral, Qwen on Your Hardware](/lukas_tanaka/local-llms-2026-run-llama-mistral-qwen-on-your-hardware-complete-guide-32k)