> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tinfoil.sh/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> This page documents significant changes to the Tinfoil API, SDKs, and documentation.

## November 2026

### Model Sunset: Kimi K3

Kimi K3 (`kimi-k3`) has been taken offline. Use [Kimi K3.1 Pro](https://docs.tinfoil.sh/models/chat.md) (`kimi-k3-1-pro`) instead.

### New Model: Kimi K3.1 Pro

Added support for Kimi K3.1 Pro, Moonshot's flagship multimodal model post-trained from Kimi K3, for long-horizon coding, agentic workflows with tool calling, financial research, and visual knowledge work, with always-on reasoning and configurable `reasoning_effort`.
Model ID: `kimi-k3-1-pro`

### Model Deprecation: Kimi K3

Kimi K3 (`kimi-k3`) is deprecated and will be removed from the catalog on November 12, 2026. Use Kimi K3.1 Pro (`kimi-k3-1-pro`) instead.

***

## October 2026

### Model Sunset: GLM-5.3 Flash

GLM-5.3 Flash (`glm-5-3-flash`) has been removed from the catalog. Use DeepSeek V4.1 Flash (`deepseek-v4-1-flash`) instead.

### Model Deprecation: GLM-5.3 Flash

GLM-5.3 Flash (`glm-5-3-flash`) is deprecated and will be removed from the catalog on October 9, 2026. Use DeepSeek V4.1 Flash (`deepseek-v4-1-flash`) instead.

***

## September 2026

### Model Sunset: DeepSeek V4 Flash

DeepSeek V4 Flash (`deepseek-v4-flash`) has been removed from the catalog. Use DeepSeek V4.1 Flash (`deepseek-v4-1-flash`) instead.

### New Model: DeepSeek V4.1 Flash (Experimental)

Added support for DeepSeek V4.1 Flash, DeepSeek's efficient MoE model with a 1M-token context window, image input, optional reasoning with configurable `reasoning_effort`, built-in speculative decoding, tool calling, and structured outputs.
Model ID: `deepseek-v4-1-flash`

### Model Sunset: GLM-5.2

GLM-5.2 (`glm-5-2`) has been removed from the catalog. Use GLM-5.3 (`glm-5-3`) instead.

### New Model: GLM-5.3 Flash

Added support for GLM-5.3 Flash, Z.AI's fast multimodal MoE model with a 1M-token context window, image input, always-on reasoning with configurable `reasoning_effort`, tool calling, and structured outputs.
Model ID: `glm-5-3-flash`

### New Model: GLM-5.3

Added support for GLM-5.3, Z.AI's latest flagship for agentic coding and long-horizon tool use, with a 1M-token context window, always-on reasoning with configurable `reasoning_effort`, tool calling, and structured outputs.
Model ID: `glm-5-3`

***

## August 2026

### Model Sunset: Qwen3-VL 30B

Qwen3-VL 30B (`qwen3-vl-30b`) has been removed from the catalog.

### New Model: DeepSeek V4 Flash

Added support for DeepSeek V4 Flash (the [0731 release](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731)), DeepSeek's efficient MoE model with a 1M-token context window, built-in speculative decoding, tool calling, and structured outputs.
Model ID: `deepseek-v4-flash`

### New Model: Kimi K3

Added support for Kimi K3, Moonshot's flagship multimodal model for complex reasoning, long-horizon coding, visual knowledge work, and agentic workflows with tool calling.
Model ID: `kimi-k3`

***

## July 2026

### Model Sunset: Kimi K2.6

Kimi K2.6 (`kimi-k2-6`) has been removed from the catalog.

### Model Sunset: DeepSeek V4 Pro

DeepSeek V4 Pro (`deepseek-v4-pro`) has been removed from the catalog.

***

## June 2026

### New Model: GLM-5.2

Added support for GLM-5.2, a frontier-class model built for agentic engineering with state-of-the-art performance on coding benchmarks and sustained reasoning over long tool-use sessions, with an extended 384K-token context window.
Model ID: `glm-5-2`

### Model Sunset: GLM-5.1

GLM-5.1 (`glm-5-1`) has been removed from the catalog. Use GLM-5.2 (`glm-5-2`) instead.

***

## May 2026

### New SDK: Rust

Released the [Tinfoil Rust SDK](https://docs.tinfoil.sh/sdk/rust-sdk.md) at v0.0.3 with attestation-verified TLS, an async-openai-compatible chat surface, and first-class shortcuts for chat (`chat_relaxed`), embeddings (`embed_batch`), and audio transcription (`transcribe`). Streams Tinfoil-specific vendor events (web search progress, structured-outputs constraints, PII checks).
Crate: `tinfoilsh/tinfoil-rs`

***

## April 2026

### Document Processing API: New Pipeline

PDF parsing moved from docling to sandboxed MuPDF; scanned pages route to the `gemma4-31b` VLM. Adds a `mode` query parameter on `/v1/convert/file` (`text`, `vision`, `images`, `raw`, `vlm`), multi-file uploads, and proper `502` failures on OCR errors. See the [Document Processing Guide](https://docs.tinfoil.sh/guides/document-processing.md).

### New Model: DeepSeek V4 Pro

Added support for DeepSeek V4 Pro, a 1.6T-parameter Mixture-of-Experts model with 49B active parameters per token, strong long-context reasoning, coding, math, and agentic task performance.
Model ID: `deepseek-v4-pro`

### New Model: Kimi K2.6

Added support for Kimi K2.6, Moonshot's latest multimodal agentic model with stronger long-horizon coding, design-to-code, image and video understanding, and large-scale tool-using workflows.
Model ID: `kimi-k2-6`

### New Model: GLM-5.1

Added support for GLM-5.1, a frontier-class model built for agentic engineering with state-of-the-art performance on coding benchmarks and sustained reasoning over long tool-use sessions.
Model ID: `glm-5-1`

### New Model: Gemma 4 31B

Added support for Gemma 4 31B, a multimodal model from Google DeepMind with image understanding, built-in thinking mode, and native function calling.
Model ID: `gemma4-31b`

### Model Sunset: DeepSeek R1

DeepSeek R1 (`deepseek-r1-0528`) has been removed from the catalog.

### Model Sunset: Kimi K2.5

Kimi K2.5 (`kimi-k2-5`) has been removed from the catalog.

***

## February 2026

### New Model: Kimi K2.5

Added support for Kimi K2.5, a multimodal model that processes images and text together. Capable of generating code from visual designs and coordinating parallel sub-agents for complex tasks.
Model ID: `kimi-k2-5`

### Model Sunset: Qwen3 Coder 480B, Kimi K2 Thinking

Qwen3 Coder 480B (`qwen3-coder-480b`) and Kimi K2 Thinking (`kimi-k2-thinking`) have been removed from the catalog.

***

## January 2026

### New Model: Voxtral Small 24B

Added support for Voxtral Small 24B, an audio-capable model built on Mistral Small 3.1 for transcription, translation, and spoken queries.
Model ID: `voxtral-small-24b`

### New Model: GPT-OSS Safeguard 120B

Added support for GPT-OSS Safeguard 120B, a safety reasoning model for content classification and Trust & Safety applications.
Model ID: `gpt-oss-safeguard-120b`

***

## December 2025

### New Model: Kimi K2 Thinking

Added support for Kimi K2 Thinking, an advanced thinking agent with deep multi-step reasoning and stable long-horizon tool orchestration capabilities.
Model ID: `kimi-k2-thinking`

### Model Sunset: DeepSeek V3.1 Terminus

DeepSeek V3.1 Terminus (`deepseek-v31-terminus`) has been removed from the catalog.

### Tinfoil Code Sunset

The Tinfoil Code experiment has been discontinued.

***

## November 2025

### New Model: Qwen3-VL 30B

Added support for Qwen3-VL 30B, an advanced vision-language model for image understanding with exceptional multimodal capabilities.
Model ID: `qwen3-vl-30b`

### Models Sunset

The following models have been removed from the catalog:

* **Mistral Small 3.1 24B** (`mistral-small-3-1-24b`)
* **Qwen 2.5 72B** (`qwen2-5-72b`)
* **Kokoro** (`kokoro`)

***

## October 2025

### Encrypted HTTP Body Protocol (EHBP)

**Breaking Change**: The API now supports Encrypted HTTP Body Protocol (EHBP) for end-to-end encryption of request and response payloads using attested encryption keys while keeping HTTP headers in the clear for routing purposes.
EHBP cryptographically guarantees that user data can only be decrypted inside the secure enclave when the key is retrieved from an attestation report.

**Migration Required**: Update all SDKs to version `0.10.0` or greater to maintain compatibility.

* Python: `pip install --upgrade tinfoil>=0.10.0`
* JavaScript: `npm install tinfoil@latest`
* Swift: Update package dependency to `0.10.0` or greater
* Go: `go get github.com/tinfoilsh/tinfoil-go@latest`

See the [EHBP documentation](https://docs.tinfoil.sh/resources/ehbp.md) for technical details.

***

## September 2025

### New Model: Qwen3 Coder 480B

Added support for Qwen3 Coder 480B, a powerful coding model with advanced agentic capabilities and understanding.
Model ID: `qwen3-coder-480b`

### Tinfoil Code Launch (Experimental)

Released Tinfoil Code, a privacy-focused fork of Qwen-Code coding agent running Qwen3 Coder 480B.

* Install: `npm install -g @tinfoil-code/qwen-code@latest`
* Run: `tinfoil-qwen`

***

## August 2025

### Admin API Launch

Introduced Admin API for programmatic user resource management.

* Endpoints for API key management
* Billing and usage statistics access

***

## July 2025

### SDK Updates - Simplified Client Initialization

All models can now be accessed through the unified model router hosted at `https://inference.tinfoil.sh`

The router automatically routes to the appropriate model enclave based on your API request
This maintains a consistent endpoint across all models and matches OpenAI API standards.

**Breaking Change**:  The `enclave` and `repo` parameters are now deprecated in the client initialization and have been removed from client constructors.

**Before:**

```python theme={"dark"}
client = TinfoilAI(
    enclave="<enclave>.model.tinfoil.sh",
    repo="tinfoilsh/<config-repo>",
    api_key=os.getenv("TINFOIL_API_KEY")
)
```

**After:**

```python theme={"dark"}
client = TinfoilAI(
    api_key=os.getenv("TINFOIL_API_KEY")
)
```

**Migration Guide:**
Remove the `enclave` and `repo` parameters from your client initialization

Keep only the API key parameter. The SDK now automatically routes requests through `https://inference.tinfoil.sh`.

***

## June 2025

***

## May 2025

### New Audio Model: Whisper Large V3 Turbo

Added support for Whisper Large V3 Turbo for speech-to-text transcription. The chat now also offers voice inputs.\
Model ID: `whisper-large-v3-turbo`

### Document Processing API Launch

Added support for secure document processing:

* Endpoint: `POST https://inference.tinfoil.sh/v1/convert/file`
* Supports PDF, DOCX, PPTX, XLSX, HTML, and images
* Advanced understanding of document structure, layout, tables, and formulas

***

## April 2025

### New Model: Llama 3.3 70B

Added support for Llama 3.3 70B, an advanced conversational AI model.
Model ID: `llama3-3-70b`

***

> [!NOTE]
>
> **Need Help?** Contact us at [support@tinfoil.sh](mailto:support@tinfoil.sh) for assistance with migrations or if you have questions about these changes.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.