Model lineup expanded. GPT-5.6, Grok 4.6, and multi-vendor access
Perplexity's August 2026 update expanded model access substantially: GPT-5.6 Terra and Luna added to Computer roles, Grok 4.6 availability expanded, and Agent API now supports xai/grok-4.6 as xAI's flagship reasoning model. Perplexity's 2026 model access splits between Pro and Max tiers; not every paid user sees the same list. Full model roster spans Perplexity's own Sonar family plus models from OpenAI, Anthropic, Google, xAI, Moonshot AI, Z.ai, and NVIDIA.
What this means for you
For practice owners using Perplexity for clinical research or content grounding: model choice matters. Sonar (Perplexity's own) is fastest but shallower; GPT-5.6 and Claude Opus 5.5 deliver richer analysis. For medical/clinical research questions, GPT-5.6 Terra tends to produce more thorough citation lists. Pro plan gets baseline access; Max is needed for full model selection.
Source Context Panel launched, citations kept visible alongside responses
Perplexity introduced the Source Context Panel, which keeps citations and supporting sources visible in a dedicated panel alongside the response. Users can inspect evidence behind each claim, compare sources, and continue research without leaving the conversation. Typical Perplexity responses now include 5-10 inline citations, with every claim linked to its source. Perplexity typically reads ~10 pages per query and cites 3-4, a much narrower filter than ChatGPT.
What this means for you
For healthcare content authors: Perplexity's brutal 10-to-3 filter means only the most extractable, well-attributed content gets cited. Playbook: publish content where each key claim has a clear inline citation to a primary source (peer-reviewed study, official guideline, verified statistic), use direct question-first paragraphs, and update timestamps regularly. Fresh, extractable, well-attributed. Perplexity's filter rewards all three simultaneously.
Search as Code + unified SDK, execution reliability up from 82% to 93%
Perplexity rolled out 'Search as Code' optimizations in Computer, routing search through a unified SDK-backed interface. Two update batches raised execution reliability from 81.9% to 92.6%. This affects any team building on the Perplexity API for automated research, competitive intelligence, or content grounding. Reliability improvements matter most at scale.
What this means for you
For practices using Perplexity API in workflows (e.g. automated research summaries for clinicians, competitive intel pulls): the reliability improvement means fewer silent failures and less need to build retry logic. Directly usable for AGM internal workflows; less directly consumer-facing.
