# Together AI — AdapTive-LeArning Speculator System (ATLAS): A New Paradigm in LLM Inference via Runtime-Learning Accelerators

- Company: Together AI (together.ai)
- Announced: 2025-10-10T00:00:00+00:00
- Category: model-update
- Subject: Inference platform
- Models affected: DeepSeek-V3.1, Kimi-K2
- Source: https://www.together.ai/blog/adaptive-learning-speculator-system-atlas
- Record: https://forck.live/items/2407-adaptive-learning-speculator-system-atlas-a-new-paradigm-in-llm-inference-via

Together AI announces ATLAS, a new adaptive-learning speculator system that improves inference performance at runtime, achieving up to 500 TPS on DeepSeek-V3.1 and up to 460 TPS on Kimi-K2.

## Evidence

Verbatim from https://www.together.ai/blog/adaptive-learning-speculator-system-atlas:

> ATLAS offers a new way of doing speculative decoding — one that dynamically improves at runtime — and it fits seamlessly alongside our other Turbo techniques like the proprietary Together Turbo Speculator or Custom Speculators.

---

Record: https://forck.live/items/2407-adaptive-learning-speculator-system-atlas-a-new-paradigm-in-llm-inference-via
Catalogue: https://forck.live/llms.txt
Feed: https://forck.live/feed.md