Promptfoo 0.122.1
Promptfoo 0.122.1 brings major tracing updates aligned with OpenTelemetry GenAI conventions, adds external trace ingestion (Grafana Tempo, Langfuse, Braintrust), introduces Gemini 3.7 Flash models, and refines red team token accounting.
Why this signal matters
Promptfoo 0.122.1 introduces major tracing and evaluation enhancements. Key tracing updates include alignment with OpenTelemetry GenAI conventions, support for universal target spans, test-case roots, standardized tool execution, telemetry integration for OpenAI Agents, Vercel SDK, and Claude/Codex subprocesses, as well as fetching external traces from Grafana Tempo, Langfuse, and Braintrust. For evaluation and red teaming, the release improves token accounting by separating scan footprints from incurred usage, detailing generation, target, attacker, and grading token counts, including multi-turn and remote strategy calls. Provider updates introduce Gemini 3.7 Flash and Flash-Lite, improve Claude token and thinking budget calculations, harden coding-agent policies, add path-traversal guards on callback loaders, and redact AWS SigV4 credentials.
Actionable summary
Promptfoo version 0.122.1 adds OpenTelemetry GenAI tracing alignment, external trace fetching (Tempo, Braintrust, Langfuse), support for Gemini 3.7 Flash/Flash-Lite, and updates red team accounting to separate target, attacker, generation, and grading tokens.
- Agent usefulness
- 62/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade Promptfoo to 0.122.1 using your package manager.
- Review OpenTelemetry GenAI trace configurations if relying on external providers like Grafana Tempo, Langfuse, or Braintrust.
- Verify token accounting dashboards or metrics, taking into account separated scan footprint and multi-turn red team token counts.