AI Agent Builders Urged to Decouple Models as OpenAI Share Falls to 27%
Updated
Updated · InfoWorld · Oct 26
AI Agent Builders Urged to Decouple Models as OpenAI Share Falls to 27%
3 articles · Updated · InfoWorld · Oct 26
Summary
Stable interfaces between agents and models, with tools and state kept outside any one provider, are being pitched as the safest design as model retirements and API shutdowns keep forcing migrations.
Anthropic retired Claude Sonnet 4 and Opus 4 from its API in 2026, and OpenAI later shut its Assistants API, showing how applications break when model IDs, hosted state or provider SDKs are embedded deep in business logic.
Market churn is accelerating: Menlo Ventures said OpenAI’s enterprise LLM API share fell from 50% in 2023 to 27% by end-2025, while Anthropic rose to 40% and Google to 21%.
Costs are moving just as fast—Stanford found GPT-3.5-level MMLU performance dropped from $20 per million tokens in November 2022 to $0.07 by October 2024—making per-task routing and regular re-evaluation more valuable.
The proposed fix is a provider-agnostic adapter layer, MCP-style tool access, self-managed state and an evaluation suite so teams can switch for cost, privacy or performance without rebuilding agents.