•Anthropic recently altered its API pricing model for programmatic access, reportedly shifting from a subscription-based approach to per-use API pricing.
•Prominent developer José Valim criticized Anthropic's communication, suggesting it was framed misleadingly as an 'upgrade' for investors and enterprises, while impacting developers negatively.
•The incident highlights the critical need for transparency, honesty, and strong developer relations from AI platform providers to maintain trust within the developer community.
•Anthropic silently reduced the cache Time-To-Live (TTL) for an unspecified API from 1 hour to 5 minutes around early March 2026.
•This change was not officially announced, leading to unexpected 'quota and cost inflation' for users relying on previous caching behavior.
•Developers and teams using Anthropic services should review their API usage patterns and billing statements from early March onwards to detect potential impacts.
•OpenAI Codex is transitioning from per-message pricing to a more granular API token usage model.
•The new model charges based on input tokens, cached input tokens, and output tokens, offering greater transparency into credit consumption.
•Different Codex models (e.g., GPT-5.4, GPT-5.4-Mini) have varying credit rates per million tokens.
•Existing Plus/Pro/Edu and Enterprise/Edu customers will temporarily remain on a legacy per-message rate card until fully migrated.
•Developers can monitor token usage in Codex settings and should optimize for token consumption to manage costs, with average costs estimated at $100-$200/developer/month.
•Anthropic recently altered its API pricing model for programmatic access, reportedly shifting from a subscription-based approach to per-use API pricing.
•Prominent developer José Valim criticized Anthropic's communication, suggesting it was framed misleadingly as an 'upgrade' for investors and enterprises, while impacting developers negatively.
•The incident highlights the critical need for transparency, honesty, and strong developer relations from AI platform providers to maintain trust within the developer community.
•Anthropic silently reduced the cache Time-To-Live (TTL) for an unspecified API from 1 hour to 5 minutes around early March 2026.
•This change was not officially announced, leading to unexpected 'quota and cost inflation' for users relying on previous caching behavior.
•Developers and teams using Anthropic services should review their API usage patterns and billing statements from early March onwards to detect potential impacts.
•OpenAI Codex is transitioning from per-message pricing to a more granular API token usage model.
•The new model charges based on input tokens, cached input tokens, and output tokens, offering greater transparency into credit consumption.
•Different Codex models (e.g., GPT-5.4, GPT-5.4-Mini) have varying credit rates per million tokens.
•Existing Plus/Pro/Edu and Enterprise/Edu customers will temporarily remain on a legacy per-message rate card until fully migrated.
•Developers can monitor token usage in Codex settings and should optimize for token consumption to manage costs, with average costs estimated at $100-$200/developer/month.
