generated: '2026-07-21'
method: searched
source: https://docs.sarvam.ai/api/getting-started/changelog
scheme: dated-monthly
notes: >-
  Sarvam publishes a dated, monthly changelog grouped by product area
  (Chat Completion, Speech to Text, Text to Speech, Pricing, Integrations).
entries:
- date: '2026-06'
  highlights:
  - product: chat
    breaking: true
    text: sarvam-m deprecated and no longer accepted; use sarvam-30b (64K) or sarvam-105b (128K).
  - product: speech-to-text
    breaking: false
    text: /speech-to-text REST and Batch now default to saaras:v3 (was saarika:v2.5); respects mode parameter.
  - product: speech-to-text
    breaking: false
    text: /speech-to-text-translate is now legacy (saaras:v2.5); use /speech-to-text with saaras:v3 and mode=translate.
- date: '2026-05'
  highlights:
  - product: chat
    breaking: true
    text: Removed fixed-context sarvam-30b-16k and sarvam-105b-32k variants; base models now serve full context.
  - product: pricing
    breaking: false
    text: Document Intelligence renamed Document Digitization and repriced to ₹0.5/page (from ₹1.5).
  - product: pricing
    breaking: false
    text: New accounts now receive ₹100 free credits (previously ₹1,000).
- date: '2026-04'
  highlights:
  - product: chat
    breaking: false
    text: Adjusted default values for reasoning_effort and max_tokens in Chat Completions.
  - product: chat
    breaking: false
    text: Marked sarvam-105b-32k and sarvam-30b-16k variants deprecated ahead of removal.
  - product: speech-to-text
    breaking: false
    text: Added Voice Activity Detection (VAD) parameters to STT WebSocket streaming.