▌
provider specs
what you need to know before integrating
| provider | pricing model | est. cost/query | latency | api pattern | citations | special capabilities |
|---|---|---|---|---|---|---|
| OpenAI o3 ↗ o3-deep-research deprecated · shutdown 2026-12-11 → gpt-5.5-pro | per-token | $1-10/query | 10-30+ min | Responses API, async background | yes (inline annotations) | MCP serverscode interpreterfile searchweb search |
| OpenAI o4-mini ↗ o4-mini-deep-research deprecated · shutdown 2026-12-11 → gpt-5.5-pro | per-token | $0.40-2.50/query | 3-10 min | Responses API, async background | yes (inline annotations) | MCP serverscode interpreterfile searchweb search |
| Perplexity Sonar ↗ sonar-deep-research | per-token + per-search | $0.30-1.30/query | 1-3 min | Chat Completions (OpenAI-compatible), synchronous | yes (inline + source URLs) | citation-focusedsearch groundingbilled per token + citation + reasoning + search |
| Gemini Deep Research ↗ deep-research-preview-04-2026 | per-token + per-search | $1-3/task | ~20 min (60 min max) | Interactions API, async (background=true) | yes (url + file citations) | google searchurl contextcode executionMCPfile searchvisualizationmultimodal |
| Gemini Deep Research Max ↗ deep-research-max-preview-04-2026 | per-token + per-search | $3-7/task | up to 60 min | Interactions API, async (background=true) | yes (url + file citations) | ~160 searches/taskgoogle searchurl contextcode executionMCPfile searchvisualization |
| Parallel Pro ↗ pro | per-request | $0.10/query | 2-10 min | Task API, async polling | yes (with excerpts) | structured JSON outputauto-schema1 of 9 tiers (lite→ultra8x) |
| Parallel Pro-Fast ↗ pro-fast | per-request | $0.10/query | 30s-5 min | Task API, async polling | yes (with excerpts) | structured JSON outputauto-schemafaster variant (lower freshness) |
| Parallel Ultra ↗ ultra | per-request | $0.30/query | 5-25 min | Task API, async polling | yes (with excerpts) | structured JSON outputauto-schemadeep research modeultra2x/4x/8x available |
| Parallel Ultra-Fast ↗ ultra-fast | per-request | $0.30/query | 1-10 min | Task API, async polling | yes (with excerpts) | structured JSON outputauto-schemafaster variant (lower freshness) |
| Valyu DeepResearch ↗ deepresearch (fast/standard/heavy/max) | per-request | $0.10-15/task | 1 min - 3 hr | REST, async (poll / webhook) | yes (title + url) | SEC/academic/clinical/patent sourcesstructured JSON outputpython sandboxbrowser automation |
| You.com Research ↗ research (lite→frontier) | per-request | $0.012-0.45/query | < 10s - 5 min | REST, synchronous | yes (with excerpts) | structured JSON outputsource control (domains/freshness)finance research variant |
| Exa Research ↗ exa-research | per-request | ~$0.02-0.15/query (usage-based) | ~1-2 min | Research API, async polling | yes (excerpts + field grounding) | neural/semantic searchstructured output w/ groundingreturns real per-query cost |
benchmarks
the RACE column is an independent academic benchmark (DeepResearch Bench, 2025 model versions). vendor scores are self-reported — note that DRACO is authored by Perplexity and Valyu's figures are its own re-run, so treat them as marketing, not neutral evals. every number links to its source. for a blind, community-run signal, see our arena leaderboard.
| provider | RACE (independent) | vendor self-reported | arena elo |
|---|---|---|---|
| Gemini Deep Research | 48.9 ↗ | 59.0% DRACODRACO (Perplexity-authored) ↗ | view rankings → |
| OpenAI Deep Research | 47.0 ↗ | 52.1% DRACODRACO (Perplexity-authored) ↗ | view rankings → |
| Perplexity Deep Research | 42.3 ↗ | 70.5% DRACODRACO (Perplexity-authored) ↗ | view rankings → |
| Grok Deeper Search(not indexed) | 40.2 ↗ | — | view rankings → |
| Parallel Ultra (8x) | — | 82% DeepSearchQAParallel (self-reported) ↗ | view rankings → |
| Valyu DeepResearch (Heavy) | — | 72.7% DRACOValyu (self-reported re-run) ↗ | view rankings → |
| You.com Research | — | 52.9% DRACOValyu (re-run) ↗ | view rankings → |
| Exa Research | — | 25.9% DRACOValyu (re-run) ↗ | view rankings → |