New reasoning_effort parameter for sonar-deep-research to control computational effort per query.
Options:
"low": Faster, simpler answers with reduced token usage"medium": Balanced approach (default)"high": Deeper, more thorough responses with increased token usageDirectly impacts reasoning tokens consumed, enabling cost control.
Fetched May 26, 2026
The DeepSeek provider now accepts thinking.type: 'adaptive' and the full reasoningEffort enum (low, medium, high, xhigh, max).
Vercel · AI SDKFast mode (research preview) now supports Claude Opus 4.7 via with the beta header, delivering significantly faster output token generation…
Anthropic · Claude PlatformThe tool is now available in the Agent API. Pull structured financial and market data — quotes, financials, earnings, analyst estimates, se…
Perplexity · SonarAgent API: New Third-Party Models — Claude Opus 4.7, GPT-5.5, and Grok 4.20 Reasoning now supported. API Key Management: Security Upgrade —…
Perplexity · Sonar