[Highlight] Contributions to LLM Vendors

[FFengIll]

We found multiple bugs in the SDKs provided by the model vendors, such as OpenAI, Anthropic, and Alibaba. This post keeps the entries we contributed back to the vendors:

Issues

• AgentScope (Anthropic Integration): A bug in content block conversion logic causes failures when parsing responses from Anthropic models. bugfix: parse anthropic response content block type logic error. agentscope-ai/agentscope#1070

• OpenAI Go SDK Compatibility Issue: The current Go SDK does not support passing extra fields, making it incompatible with non-OpenAI-native models such as DeepSeek and Gemini. Its behavior is inconsistent with the Python and TypeScript SDKs, breaking cross-language SDK parity. feat: serialize extra fields when sending message requests openai/openai-go#593

• Anthropic Go SDK (Beta) Serialization Bug: During message serialization and deserialization, certain fields are lost, compromising the integrity of requests and responses. bugfix: missing custom decoder for beta content block which will cause field missing. anthropics/anthropic-sdk-go#263

[FFengIll]

Tingly-Box 帮助修复模型厂商SDKs汇总

• AgentScope（Anthropic 集成）：内容块（content block）转换逻辑存在缺陷，导致 Anthropic 模型响应解析失败。

• OpenAI Go SDK 兼容性问题：当前 Go SDK 不支持传递额外字段（extra fields），导致无法兼容 DeepSeek、Gemini 等非 OpenAI 原生模型。 其行为与 Python 和 TypeScript SDK 不一致，破坏了多语言 SDK 的一致性体验。

• Anthropic Go SDK（Beta 版）序列化问题 ：在消息（message）的序列化与反序列化过程中存在字段丢失的 Bug，影响请求与响应的完整性。