arize-phoenix: v20.21.0
Arize Phoenix v20.21.0 addresses CVE-2026-107286 by upgrading pydantic-ai, adds Mistral AI cost tracking, and fixes evaluator metadata drops.
Why this signal matters
Arize Phoenix has released version 20.21.0. Key updates include a security fix upgrading pydantic-ai to address CVE-2026-107286. New features introduce support for Mistral AI model pricing in cost-tracking and assign random gradients to projects created via ingestion. Other fixes include synchronizing built-in model prices and lists in the playground, resolving an issue where evaluator metadata nested under results was dropped, and bumping arize-phoenix-evals to version 3.9.2. Documentation was also restructured around agent workflows.
Actionable summary
Arize Phoenix release v20.21.0 upgrades pydantic-ai to mitigate CVE-2026-107286, integrates Mistral AI model cost tracking, updates arize-phoenix-evals to 3.9.2, and prevents dropping of nested evaluator metadata.
- Agent usefulness
- 96/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade arize-phoenix to v20.21.0 to address CVE-2026-107286 in pydantic-ai.
- Verify Mistral AI pricing tracking if using Mistral models with Phoenix cost tracking.
- Update dependency configurations for arize-phoenix-evals 3.9.2 if managed separately.