How to add web search to Kimi K3 and other open-weight models

Moonshot AI’s Kimi K3 is a 2.8-trillion-parameter mixture-of-experts model with tool calling and a one-million-token context window, but its built-in $web_search tool is scheduled to retire on October 20, 2026, requiring developers to connect external search services for current information. The discussion compares web-search availability across Kimi K3, Z.ai’s GLM 5.3, DeepSeek V4.1, and Reflection Beam, noting that tool calling is broadly supported while native search varies by provider and deployment host. It presents a provider-neutral Python function-calling loop using the OpenAI SDK and Parallel Search API, emphasizing that complete assistant responses, including reasoning fields, must be preserved between turns for models such as Kimi K3 and GLM 5.3. The example limits search-result payload sizes to reduce token costs and latency, while alternatives include OpenRouter’s server-side web search tool and a free Parallel Search MCP server for compatible clients. Testing described in the piece found that Parallel searches returned results in under a second, although the full Kimi K3 loop was validated through a local reasoning-model stub rather than with a live Moonshot or OpenRouter model connection.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.