# Release Notes — v1.2 (May 2026)

## Dataset Updates

### Prompt ID Nomenclature

* Updated Prompt ID naming convention for improved categorization and traceability.
* Refer to the Internal Categories documentation for the new nomenclature structure.

### Expanded Prompt Dataset

The dataset was upgraded with additional evaluation prompts across the following categories:

| Category               | Total Prompts |
| ---------------------- | ------------- |
| Language & Grammar     | 29            |
| Cultural Adaptation    | 34            |
| Instruction & Response | 32            |
| Multimodal             | 30            |

### New Placeholder Types Added

The following placeholder categories were introduced to improve localization and cultural evaluation coverage:

* Public Figure
* Demonym
* Diminutive
* Holiday
* Commemoration
* Memes
* Event
* Internet Slang
* Brand
* Relationship

---

## Model Updates

### Language & Grammar Expansion

* Added a new subcategory:

  * **Semantic Accuracy**

This new evaluation area improves the model’s ability to assess meaning preservation, contextual correctness, and semantic consistency across generated responses.

---

## Evaluation Updates

### Expanded Evaluation Coverage

* Increased evaluation samples from **10 prompts** to **15 prompts per subcategory** for more reliable benchmarking and consistency analysis.

### Evaluated Models

The following models were benchmarked during the v1.2 evaluation cycle:

* OpenAI ChatGPT v5.4
* DeepSeek DeepSeek v3
* Google Gemini v3 Flash
* Microsoft Copilot v5.5