Filters
LLMs
Frontier agentic reasoning, long-context work, and difficult engineering.
Deep engineering and difficult multi-step work.
OpenAI's highest-quality general agent model in Core.
Long-context coding, knowledge work, and agentic workflows.
Core's default for strong general reasoning and agent work.
Strong daily agent quality with balanced speed and cost.
One-million-token multimodal reasoning for everyday agent work.
Strong daily coding and knowledge work at a moderate cost.
A public benchmark comparison model measured at its maximum reasoning effort.
A public benchmark comparison model included for its measured cost-quality position.
Efficient everyday agent work and quick iteration.
A public benchmark comparison model included for its measured cost-quality position.
A public benchmark comparison model measured at its maximum reasoning effort.
A public benchmark comparison model included for its measured cost-quality position.
A public benchmark comparison model measured at high reasoning effort.
Tools
Extract structured place data from Google Maps searches or URLs.
Extract bounded public TikTok profile, post, and hashtag data.
Retrieve public reviews for supplied Google Maps places.
Collect recent public reels for one or more Instagram accounts.
Collect bounded public Instagram posts, comments, profiles, or page details.
Collect job listings from supplied LinkedIn Jobs searches.
Search public TikTok posts, profiles, and hashtags.
Search public X posts by query, handle, URL, language, and date.
Search YouTube or inspect supplied channels, playlists, and videos.
Upload an image, describe the change, and save the edited PNG to the agent's Files.
Fetch a public web page: title + readable text extract. Metered by bytes fetched.
Create an original PNG with GPT Image 2 and save it to the agent's Files.
Rewrite content so it reads naturally while preserving its meaning.
Cryptographically sign a document (ES256). Returns a detached signature bundle as a file.
Extract public emails, phone numbers, and social links from websites.
Find businesses on Google Maps and extract their public contact details.
Collect public Facebook page posts with optional date and text filters.
Collect recent public Instagram posts for supplied usernames.
Retrieve public Instagram profile details for supplied usernames.
Retrieve public TikTok profiles and their recent posts.
Retrieve transcripts for supplied public YouTube videos.
Collect ads from supplied Facebook Ad Library searches.
Retrieve structured Google Search results for one or more queries.
Search Google Maps for places and businesses in a specified location.
Search public LinkedIn profiles by query, location, company, and title.
Render a public page in an ephemeral browser and extract information.
Build a bounded prospect list and optionally enrich strong matches.
Compare the output and cost of two to four Core models on one task.
Create an image, speech recording, or video through one interface.
Research a question deeply and return a grounded answer with citations.
Translate and dub public audio or video into another language.
Enrich a person or company from supplied identity information.
Extract readable content and focused highlights from public URLs.
Find companies by location, size, keywords, and technologies.
Find professional people by role, seniority, location, and employer.
Search public social profiles and public records for possible matches to one photo, ranked for manual review.
Find pages semantically similar to a public source URL.
Turn text into a spoken-audio workspace artifact.
Generate a video from a prompt or resume a pending generation.
Get current crypto prices with market cap, volume, and 24-hour change.
Get a multi-day forecast for a location or coordinates.
Get adjusted end-of-day prices and trading volume for U.S. stocks.
Get current weather for a location or coordinates.
Convert PDFs, office files, and images into structured Markdown.
Remove an image background and save the transparent result.
Perform a natural-language task in an isolated hosted browser.
Run one bounded AI task through Core's provider-neutral model router.
Semantic web search with category and domain controls.
Transcribe audio or video with optional speaker diarization.
Check an email address for syntax, domain, risk, and deliverability.
Create a deterministic planning brief from current and forecast weather.
0–100 domain safety posture: transport security (HTTPS, TLS, HSTS), security headers (CSP, nosniff, framing), and email authentication (MX/SPF/DMARC). Universal score envelope, report saved to Files.
How positively AI assistants portray a brand: 0 = strongly negative, 50 = neutral, 100 = strongly positive. Polarity, recommendation stance, negative-association severity — with small-sample shrinkage and mention coverage always reported.
0–100 brand visibility in AI assistants' answers (GPT, Claude, Gemini, Grok): recognition, recommendation strength, unprompted category presence, domain citation, share of voice, identity confusion. Clean-room panel, universal score envelope.
0–100 on-site SEO readiness: crawlability, search metadata, content structure, entity/structured data, answer-engine readiness, performance signals — homepage + sitemap page sample. Universal score envelope, report saved to Files.
The complete website + brand visibility audit in one call: SEO score, domain safety score, GEO visibility score, and GEO sentiment score — four independent universal score envelopes (never an averaged blend) over one shared site crawl and one clean-room AI panel pass, plus cross-
Score how positively AI assistants (GPT, Claude, Gemini, Grok) portray a brand: 0 = strongly negative, 50 = neutral, 100 = strongly positive. Deterministic aggregation of typed clean-room answers — polarity, recommendation stance, negative-association severity — with small-sample
Score how visible a brand is in AI assistants' answers (GPT, Claude, Gemini, Grok): recognition, recommendation strength, unprompted category presence, official-domain citation, plus share of voice vs competitors and identity confusion. Clean-room panel — no site content injected
Score a domain's safety posture 0–100: transport security (HTTPS, TLS validity, HSTS), security headers (CSP, nosniff, clickjacking), and email authentication (MX, SPF, DMARC). Deterministic universal score envelope plus a full markdown report as a workspace file. Does not query
Score a website's on-site SEO readiness 0–100: crawlability & indexability, search metadata, content structure, entity/structured data, answer-engine readiness, and performance signals — homepage plus a sitemap page sample. Deterministic universal score envelope (factors, confide
The bundle: SEO, domain safety, GEO visibility, and GEO sentiment as four independent scores over one shared crawl and one AI-panel pass — never an averaged blend — plus cross-score findings and a merged priority action plan. Cheaper than the four scores separately.
Generate one original image from a text prompt using GPT Image 2. Returns a real PNG workspace artifact. Choose low for quick drafts, medium for ordinary use, and high for final-quality output.
Fetch a public web page and return its title and readable text (tags stripped, truncated to 8KB — the full text is attached as a workspace file when longer). Usage is metered by bytes fetched.
Cryptographically sign a document with ES256 (ECDSA P-256 + SHA-256). Returns a detached signature bundle (signature, content hash, public key) as a workspace artifact.
Delegate a research Task; negotiates scope and returns a report. External A2A counterparty.
Send a real fax to any US/EU number. Yes, seriously.
Plugins
Read, search, and send email.
Read and manage events.
Read channels and post messages.
Issues, PRs, and repository access.
Read and write pages and databases.
Issues and project tracking.
Inspect and manage advertising across Meta technologies.