Research / search / grounded generation

Perplexity

  • Pro monthly $20

    Model choice: Sonar 2, GPT‑5.6 Terra, Claude Sonnet 5, Gemini 3.1 Pro, Grok 4.6.

  • Max monthly $200

    Adds GPT‑5.6 Sol, Claude Opus 5 and 10,000 Computer credits a month.

  • Agent API from 09/27/2026 provider rates

    Replaces Sonar endpoints. Search, URL fetch, sandbox and MCP.

Computer on Pro and Max includes GPT‑6 Astra from 09/05/2026. Enterprise Pro is $40, Enterprise Max $325 per seat.

  • Verified2026-09-20
  • Models covered4
  • Sources6
  • Deployment channels4

Purchase decision (when to choose / when to avoid)

Choose if...

  • You do research and care about sources/citations (grounded generation).
  • You want to quickly verify trends/competition and build marketing briefs.
  • You're building search+answer pipelines or agents via the Agent API (presets fast to xhigh, sandbox, MCP).

Avoid if...

  • You need a 'pure' LLM without search and with full control over behavior/agents.
  • You have use cases where internet sources are a risk (compliance/PII) - require restrictions and policies.
  • You run an integration on the Sonar API and have no migration planned before 09/27/2026.

Cost in practice (scenarios)

Marketer/strategist (SaaS)

Pro $20/mo or Max $200/mo (Opus 5, GPT-5.6 Sol, Model Council, 10,000 Computer credits).

  • briefs, competitive analysis, trends
Team (Enterprise)

Enterprise Pro $40/seat/mo (500 Computer credits, SSO); Enterprise Max $325/seat/mo (15,000 credits, SCIM, audit logs).

  • limits: Ent Pro 400 Pro Searches/week, Ent Max 4,000/week
API (Agent API)

Tokens at provider rates (e.g. claude-sonnet-5 $2/$10, gpt-5.6-luna $0.20/$1.20) + web search $0.0025/call; low preset about $0.05 per query; Search API $5 per 1,000 requests.

  • RAG grounded on web
  • migration off Sonar before 09/27/2026
These are estimates/scenarios (not an invoice). Actual cost depends on context length, number of users, limits and retention policies.

Deployment / data / enterprise

Deployment channels

  • Perplexity (web, apps) - research for teams
  • Perplexity Computer - agent in the app, Slack, Microsoft 365 and email
  • Comet - browser with an assistant (Mac, Windows, iOS, Android)
  • Agent API and Search API - search+answer and agents in applications (Sonar until 09/27/2026)

Data policy

Training on data
Pro/Max: training opt-out in settings; Enterprise Pro and Max: no training on data; API: prompts are not logged or used for training.
Retention
Enterprise Max: configurable retention; API: zero data retention.
Data residency
Not published; require it in the enterprise agreement.
Key: source usage policies + compliance for sensitive data.

Enterprise readiness

Admin
Enterprise Pro: central management and billing; Enterprise Max: Insights dashboard.
SSO/SCIM
SSO in Enterprise Pro; SCIM from Enterprise Max.
Audit
Audit logs from Enterprise Max.
DPA
Depends on agreement.
Certifications
Declared: SOC 2 Type II, HIPAA, GDPR, PCI DSS.
Worth it when research with citations and verifiability matters.

Best use cases

  • fast desk research (benchmarks, market trends, tools, case studies)
  • competitive analysis with verifiable sources and citations
  • search+answer pipelines and agents built on the Agent API (grounded generation from the internet, sandbox, MCP).

Strengths

  • Grounded generation: answers with source list and citations - verifiability out-of-the-box.
  • Pro: Sonar 2, GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.1 Pro, Grok 4.6, Kimi K3, GLM 5.2; Max: GPT-5.6 Sol, Claude Opus 5, Model Council.
  • Computer (agent) in Slack, Microsoft 365 and email; Agent API with search, URL fetch, sandbox and MCP tools.

Weaknesses / risks

  • No fine-tuning; third-party models only through Perplexity presets and catalog, not your own deployments.
  • Sonar, Sonar Pro, Sonar Reasoning Pro and Sonar Deep Research retire on 09/27/2026 - integrations must migrate to the Agent API.
  • Less control over the base model; Computer credits and Research limits depend on plan.
  • The model picker has no Gemini Flash variants (including 3.8); the only Gemini in Pro is 3.1 Pro.

Current models (examples)

  • Sonar 2 (in-house) plus in Pro: GPT-5.6 Terra, Claude Sonnet 5, Gemini 3.1 Pro (no Flash line), Grok 4.6, Kimi K3, GLM 5.2, Nemotron 3 Ultra.
  • Max and Enterprise Max only: GPT-5.6 Sol, Claude Opus 5; Model Council (several models at once). Computer (Pro/Max): GPT-6 Astra from 09/05/2026.
  • Agent API: presets fast/low/medium (gpt-5.6-luna), high/xhigh/wide-research (gpt-5.6-sol) plus a catalog of Anthropic, OpenAI, Google, xAI, DeepSeek, Moonshot and NVIDIA models at provider rates.
  • Sonar, Sonar Pro, Sonar Reasoning Pro, Sonar Deep Research: retired 09/27/2026 (mapped to presets fast/low/medium/high).

Alternatives (if this model doesn't fit)

GR
Grok
long context and compatible API for agents
OP
OpenAI
fuller LLM API + tools; search as a module
GE
Gemini
grounding and integrations in Google ecosystem (depending on service)