Wed, July 820:22ToolsModel releases API & pricing MCP & Skills Model releases guide

promptfoo 0.121.18: Adds Grok 4.3, Claude Sonnet 5, GPT-5.6 Preview

Decision Brief

What changedpromptfoo 0.121.18 adds multiple model support, Node.js 20 support deadline, per-test repeat option, and various bug fixes.

Why it mattersDevelopers using promptfoo for evaluation or red teaming can now integrate latest models like Grok 4.3, Claude Sonnet 5, GPT-5.6, with YAML error location hints and WebSocket subprotocol support.

Who should careAI coding tool users, Agent builders

Affected stackMCPClaudeOpenAIGemini

Builder actionUpgrade MCP to 0.121.18

Source confidenceHigh · Official release / blog / repo

promptfoo 0.121.18 was released on 2026-07-07 with key features: Node.js 20 support deadline (#9862), per-test repeat option (#9781), WebSocket subprotocol display (#9947), and YAML source location hints for prompt reference errors (#9940). For model providers, it adds Bedrock support for Grok 4.3, GLM, MiniMax, Kimi, Nemotron, Gemma, Palmyra (#9788), Claude Sonnet 5 across Anthropic, Bedrock, Vertex, and Azure (#9924), GPT-5.6 preview (#9893), Moonshot (Kimi) provider (#9672), and expanded Google models (Nano Banana 2 Lite, GA image aliases, gemini-pro-latest) (#9923). Bug fixes include: is-sql not flagging SELECT DISTINCT, is-xml rejecting non-well-formed XML, is-refusal respecting inverse when output missing, CSV parsing preserving commas in quotes and zero thresholds, and Azure provider cost reporting and token refresh. Also fixed Mistral, Claude, xAI pricing and updated dependencies.

Summary basis: official / RSS sourceCompiled from the source scope noted above; the original remains authoritative.

Sources

promptfoo（评测 / 红队）
LLM eval and red-teaming toolkit — releases covering safety, jailbreak, and guardrail testing.
promptfoo（评测 / 红队）

Related intel

留言

登入后即可留言，和其他 builder 交换实测心得。

还没有留言，抢头香。