#1387·khoj

Default Anthropic models are retired - several OpenAI defaults shut down Oct 23

Author: benminorCreated Jul 30, 2026Updated Aug 21, 2026
Labelsfix

Server

  • Cloud (https://app.khoj.dev)
  • Self-Hosted Docker
  • Self-Hosted Python package
  • Self-Hosted source code

Clients

  • Web browser
  • Desktop/mobile app
  • Obsidian
  • Emacs
  • WhatsApp

OS

  • Windows
  • macOS
  • Linux
  • Android
  • iOS

Khoj version

latest

Describe the bug

Khoj's default model lists reference Anthropic models that have already been retired, so Anthropic requests on default settings return API errors — no user misconfiguration required. Anthropic retired claude-sonnet-4 / claude-opus-4 (both -0 aliases and -20250514 snapshots) on June 15, 2026, and claude-3-7-sonnet on April 20, 2026: https://platform.claude.com/docs/en/about-claude/model-deprecations

Affected references: src/khoj/utils/constants.py:16, 66, 67, 69, 70 src/khoj/processor/conversation/utils.py:93–96 src/khoj/processor/operator/__init__.py:236 src/khoj/processor/operator/operator_agent_anthropic.py:527, 545

Related but not yet failing: gpt-4.1-nano, o1, o3-mini, and o4-mini (constants.py:14/37/39/40/43, helpers.py:994, conversation/utils.py:67–73) shut down October 23, 2026 per https://platform.openai.com/docs/deprecations — flagging now since they're in the same default lists.

Current Behavior

Anthropic requests on Khoj's default model settings fail with model-not-found/deprecation errors. The defaults reference claude-sonnet-4 / claude-opus-4 (-0 aliases and -20250514 snapshots, retired June 15, 2026) and claude-3-7-sonnet (retired April 20, 2026) — https://platform.claude.com/docs/en/about-claude/model-deprecations.

References in source:

src/khoj/utils/constants.py:16, 66, 67, 69, 70 src/khoj/processor/conversation/utils.py:93–96 src/khoj/processor/operator/__init__.py:236 src/khoj/processor/operator/operator_agent_anthropic.py:527, 545

Expected Behavior

Update to latest models.

Reproduction Steps

Default model lists point at currently-served models, so Anthropic chat (and the operator agent) work out of the box with just an API key. Vendor model retirements get caught and rolled forward before the affected paths start erroring.

Possible Workaround

No response

Additional Information

I used arol-ai to find this: npx arol-ai scan

Link to Discord or Github discussion

No response