# 3.15 Launched 🚀

**URL:** <https://forum.rasa.com/t/3-15-launched/68440>\
**Category:** Announcements\
**Created:** [January 7, 2026, 8:43am UTC](https://forum.rasa.com/t/3-15-launched/68440 "2026-01-07T08:43:42Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Lauren-Goerz](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/lauren-goerz/32/20242_2.png) [@Lauren-Goerz](https://forum.rasa.com/u/Lauren-Goerz)\
**Post date:** [January 7, 2026, 8:43am UTC](https://forum.rasa.com/t/3-15-launched/68440/1 "2026-01-07T08:43:42Z")

</div>

## Rasa 3.15 is here: Observability, Voice, and Testing Power-ups!

Hey everyone,

We’re back with **Rasa 3.15**! We’ve focused on three core pillars: giving you better visibility into LLM behavior, making voice interactions feel natural, and shrinking your testing cycles.

Here’s the breakdown of what’s landing in **3.15** :

🔍 LLM Observability with Langfuse

Stop guessing how your LLMs are performing. With our new **[Langfuse integration](https://rasa.com/docs/reference/integrations/langfuse/)**, every LLM call is automatically traced.

- **Full Visibility:** Track inputs, outputs, latency, and cost in real-time.

- **Data-Driven Evaluation:** Use production traces to run evaluations and refine your prompts based on how users _actually_ talk to your bot.

### 🎙 First-Class Voice Experience

We’re closing the gap between “robotic” IVRs and human-like voice AI:

- [\* **DTMF Support**](https://rasa.com/docs/reference/primitives/flow-steps/#requesting-dtmf-input) **:** Your assistant can now handle “Press 1 for sales” or keypad entries for account numbers, perfect for noisy environments where voice recognition might struggle.

- [\* **Streaming Responses**](https://rasa.com/docs/reference/changelogs/rasa-pro-changelog/#improvements) **:** No more awkward silences. We now stream generated responses from enterprise search or the rephraser directly to your voice channels. The audio starts as soon as the first tokens arrive.

### 🧪 Ship Faster with Smarter Testing

We’ve overhauled the QA workflow so you can iterate without the wait:

- **Generative E2E Testing:** Test custom actions against “ground truth” using new [\* grounding and relevance assertions](https://rasa.com/docs/pro/testing/evaluating-assistant/#3-using-assertions-instead-of-steps).

- **Isolate & Re-run Failures:** Don’t wait for your entire suite to finish. [\* Export only the failed tests](https://rasa.com/docs/reference/api/command-line-interface/#rasa-test-e2e) and re-run them specifically:

### ✨ The “Quality of Life” List

- **DateTime Awareness:** Rasa now natively understands “quarter past 9, two days after yesterday” out of the box. See adjustments in the [\* new system prompt template](https://rasa.com/docs/reference/config/components/llm-command-generators/#prompt-template).

- **Clarification Control:** You can now limit the [\* number of options](https://rasa.com/docs/reference/primitives/patterns/#configuring-the-maximum-number-of-clarification-options) presented in clarification patterns to avoid overwhelming users.

Ready to upgrade?

Check out the full [changelog](https://rasa.com/docs/reference/changelogs/rasa-pro-changelog) for the breakdown, and don’t forget to try the [Rasa Playground](https://hellorasa.info/4stjgr5) to try out Rasa quickly and easily.