Sonar

Agent API low, medium, and high presets now use Fast Search (search_type: "fast") for web_search, dropping cost from $2.50 to $1.00 per 1,000 invocations and about 800 ms faster. xhigh is unchanged.

The Agent API now accepts PDF, DOC, DOCX, TXT, and RTF documents as input_file content via inline file_data or public HTTPS file_url. Conversations can continue with previous_response_id without re-uploading inline documents.

Agent API now supports google/gemini-4-argon, with introductory pricing through Nov 30, 2026 of $2.00 per million input tokens, $0.10 per million cached input tokens, and $10.00 per million output tokens including reasoning.

pplx-decider-v1.2 replaces v1.1-27b as the 27B decision model, and new pplx-decider-flash-v1.2 (4B) arrives at $0.01/M input, with output tokens free. Both accept text and images, and the pplx-decider-latest and pplx-decider-flash-latest aliases are added.

The Agent API xhigh preset now allows up to 512,000 output tokens, up from 128,000, and frozen configurations should update max_output_tokens to keep matching the preset. The notes also correct that xhigh has used Fast Search since the October update.

Several Anthropic, Google, and xAI models such as claude-opus-4-x, gemini-3-flash-preview, and grok-4.3 were removed from Agent API model docs, the pricing calculator, and examples, with anthropic/claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5, google/gemini-4-argon, google/gemini-3.8-flash, and xai/grok-4.7 given as the new recommended IDs. openai/gpt-6-astra and perplexity/deepseek-v4-pro-0813 were also added and are available on the Orchestrator API.

Agent API responses now include token usage and cost for successful pplx_sdk.llm.extract calls inside the sandbox tool, reported in usage.tool_calls_details.sandbox and included in usage.cost.total_cost. Uses openai/gpt-6-luna pricing. Does not backfill historical…

New image_search Agent API tool returns typed image_search_results with image and source-page URLs, with result limits and domain, format, and safe-search filters. Costs $2.50 per 1,000 successful invocations.

Setting require_approval on mcp or connector tools pauses calls for review, returning an mcp_approval_request item that is approved or denied with an mcp_approval_response using previous_response_id. Python SDK 0.43.8 and TypeScript SDK 0.38.8 include the approval types.

Agent API and Orchestrator API now support perplexity/nemotron-3.5-super-120b-a12b at $0.20/M input, $0.04/M cached input, and $0.50/M output, with a 1M-token context window. It replaces nemotron-3-ultra-550b-a55b.

pplx-decider-v1.1-27b replaces pplx-decider-v1-27b as the Decisions API model. Input now costs $0.02 per million tokens, down from $0.04. Old model name continues to work at the same rate.

The Agent API now supports anthropic/claude-haiku-5-5. Requests with more than 100k input tokens use higher rates.

Agent API now supports google/gemini-3.8-flash with promotional pricing through December 31, 2026.

Agent API now supports openai/gpt-6-sol, openai/gpt-6-luna, anthropic/claude-opus-5-5, and xai/grok-4.7.

The Agent API now supports anthropic/claude-sonnet-5-5.

Agent API low and medium presets moved to openai/gpt-6-luna; high moved to openai/gpt-6-sol.

Agent API and Orchestrator API now support perplexity/glm-5.3-flash.

On October 24, 2026, the Agent API and Orchestrator API will retire openai/gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.2, gpt-5.1, gpt-5, and gpt-5-mini. Update model selections and fallback chains before the cutoff.

Agent API fast preset now uses Fast Search (search_type: "fast"), dropping web_search cost from $2.50 to $1.00 per 1,000 invocations.

Set search_type: "fast" on the Search API or Agent API web_search tool for lower-latency search at $1.00 per 1,000 requests or invocations. Set search_type: "web" for standard search.