Here's what we shipped this week.

Updated landing page playground

The easiest way for you to try Interfaze is right on the landing page for FREE without signing up.

Try here: https://interfaze.ai

The First Open Source Diffusion Audio ASR Model 15x faster than Whisper

We achieved this by fusing DiffusionGemma and Whisper small encoders by denoising an entire transcript in parallel

Full breakdown: https://interfaze.ai/blog/the-first-open-source-diffusion-audio-asr-model

Demo: https://huggingface.co/spaces/interfaze-ai/diffusion-gemma-asr-demo

Model weights: https://huggingface.co/interfaze-ai/diffusion-gemma-asr-small

Claude Sonnet 5 benchmarks

While it's a huge improvement over 4.6 it still underperforms across all 9 benchmarks for OCR, Object detection, text-to-sql and structured output while maintaining it's high cost.

Check out full open benchmarks: https://interfaze.ai/leaderboards

Mistral OCR 4 benchmarks

Mitral OCR 4 underperforms against Interfaze and Chandra OCR 2 across a wide variety of tasks from PDF scans to multilingual support.

Full open benchmarks: https://interfaze.ai/leaderboards/olmocr

Interfaze skills

Use Interfaze skills with any tool that support it like Codex, Claude Code and more.

Repo: https://github.com/InterfazeAI/interfaze-skills

Coming soon:

• Public keys for client side use
• Improve GUI detection performance
• Better deep search capabilities
• Improved ASR mixed language audio
• Reduced token volume per request

That's it for this week!