ChatGPT (web and app, GPT-5.5)
9/10The default chat product, now powered by GPT-5.5 Instant as of April 2026 with GPT-5.5 Thinking as the reasoning variant. It has memory across conversations, file upload, image input, web browsing, and integrates Sora video, image generation, voice, and the agent tools in one interface. The interface has consolidated; Canvas was retired in GPT-5.5 in favour of inline writing and code blocks.
Key features
GPT-5.5 Instant default model with materially better STEM, web search, and concisenessPersistent memory across chats with manual override and project-scoped memoryProjects (workspaces with shared files and instructions, similar to Claude Projects)Native multimodal input including PDFs, images, audio, and screen sharingInline writing blocks and code blocks replacing the separate Canvas surface
Best use cases
- Daily drafting of emails, proposals, and Telegram-style short copy in one tab
- Photographing a contractor quote and asking for a clean comparison table
- Translating site documents between English, Polish, and Romanian on a phone
- Writing first-draft job adverts and interview questions for ops roles
- Quick UK regulation lookups with the built-in browse tool
Weaknesses
Long-context comprehension is still behind Claude past 80K tokens; independent testing puts Claude five to six points ahead on long-document retrieval. Memory occasionally pollutes unrelated chats. The Pro 1M-token context is real but the model gets lazy on dense source packs. Hallucinations on UK-specific regulation are more common than Perplexity.
Pricing
Free tier with GPT-5.3 Instant and a 10-message-per-5-hours cap. Go GBP 8 per month, Plus GBP 20 per month with GPT-5.5, Pro GBP 100 per month (5x limits) or GBP 200 per month (20x limits, 1M-token context). Team GBP 25 per user per month annual. Enterprise from circa GBP 60 per user per month with a 150-seat floor.
The universal generalist; the safest single AI to put in front of an SME team.
Configurable, file-aware versions of ChatGPT that you build once and reuse or share. You set the system prompt, attach knowledge files, switch on tools (browse, image, code), and optionally wire up Actions that call your own APIs. The store remains the only mainstream way to distribute a configured AI assistant to staff or customers without building software.
Key features
System prompt, knowledge files, and tool toggles in one builder UIActions for calling external APIs from within a conversationShareable via link inside Team and Enterprise; publishable in the GPT storeBuilt on GPT-5.4 platform with optional GPT-5.5 upgrades for paid tiersInherits ChatGPT's image, voice, and code tools automatically
Best use cases
- Bid response assistant pre-loaded with your capability statements and case studies
- Onboarding GPT for new starters with HR policies and SOPs attached
- Sales qualifier GPT that scores inbound leads against your ICP
- Customer support triage GPT that drafts replies in your brand voice
- Internal compliance GPT with the relevant UK guidance attached
Weaknesses
Context across sessions is shallow compared with Claude Projects; a Custom GPT forgets the thread of a long deal between visits. Actions are powerful but the build experience is fiddly and undocumented in places. Public store discoverability is poor, so it is not really a distribution channel for SMEs.
Pricing
Included with Plus, Pro, Team, and Enterprise. Building is free; usage counts against the underlying plan's message limits. Sharing publicly requires Plus or above; sharing inside a tenant requires Team or Enterprise.
Useful as an internal assistant template; do not mistake the store for a real distribution channel.
ChatGPT Agent (formerly Operator)
6/10The consolidated agentic mode that merges the old Operator browser, Deep Research, and the conversational model into one. It runs in a sandboxed virtual machine with a visual browser, text browser, terminal, and file access, can connect to Gmail and GitHub, fills forms, edits spreadsheets, and pauses for credentials or consequential actions.
Key features
Sandboxed VM with visual browser plus terminal you can watch liveConnectors for Gmail, GitHub, and document repositoriesPauses before consequential actions like sending email or making purchasesReal-time narration of reasoning steps with interrupt and takeoverStrong on multi-step research and analyst-style modelling tasks
Best use cases
- Pulling a competitor scan across ten UK supplier websites into one table
- Running a recurring weekly check on Companies House for named directors
- Filling out repetitive supplier onboarding forms from a master record
- Building a draft three-statement model from a target's filed accounts
- Booking travel from a brief and a corporate card without intervention
Weaknesses
Still slow and brittle on real-world sites with cookie banners, anti-bot challenges, or non-standard layouts. Plus quota burns fast. Claude's Computer Use is now competitive on developer-shaped tasks. For high-volume scraping or structured automation, n8n plus a dedicated scraping service still wins.
Pricing
Plus subscribers get circa 30 to 40 agent uses per month. Pro gets materially more capacity. Not available on Free or Go.
Worth experimenting with for repetitive web work; not yet reliable enough for unattended runs.
Advanced Data Analysis (Code Interpreter)
9/10A Python sandbox inside ChatGPT that runs code on uploaded files. It handles CSV, Excel, PDF, image, audio, and video inputs, and can produce charts, cleaned datasets, and PowerPoint or Word outputs. For most non-developer SME teams this is the single most useful feature in the product.
Key features
Python execution on uploaded files with libraries pre-installedMulti-file handling including zipped folders and large CSVsChart generation with matplotlib and basic interactive HTMLPDF and Excel parsing including merged cells and odd layoutsDirect export to Excel, PowerPoint, and Word
Best use cases
- Cleaning a messy CRM export of UK leads and deduplicating against Companies House
- Turning a year of Xero P and L exports into a single trend chart
- Reconciling supplier invoices against a purchase order log
- Generating a sales pipeline summary deck from a Salesforce export
- Parsing scanned contractor quote PDFs into a comparison spreadsheet
Weaknesses
Sessions time out and the sandbox sometimes forgets earlier state mid-thread. Large Excel files with macros or external links break. No live connection to a database, so you are always working from a snapshot.
Pricing
Included from Plus upward. Not available on Free.
The single most useful ChatGPT feature for a non-developer ops team.
GPT Image 2 (image generation)
8/10OpenAI's current image model, launched April 2026, replacing DALL-E 3 which was sunset in late 2025. Renders accurate text inside images (circa 99 percent character accuracy), handles compositional editing, multi-object scenes, and product-style outputs better than any consumer competitor.
Key features
Accurate in-image text rendering across languagesCompositional editing: change one element without re-rolling the sceneStrong product, packaging, and infographic outputsConversational refinement inside the same ChatGPT threadAPI access at circa USD 0.04 per image upward
Best use cases
- Mocking up branded sales collateral with the company name spelled correctly
- Generating a product hero shot from a description plus a reference photo
- Creating LinkedIn post imagery in a consistent house style
- Drafting infographic-style explainers for a client briefing
- Iterating signage and packaging concepts before sending to a designer
Weaknesses
Midjourney V8 still wins on pure aesthetic quality, especially photoreal portraits and cinematic scenes. GPT Image 2 looks correct but rarely beautiful. No public moodboards or style references in the way Midjourney offers.
Pricing
Included with Plus, Pro, Team, Enterprise; daily caps apply on Plus. API priced per image, from USD 0.04.
The right image model when the copy on the image has to be spelled correctly.
Native speech-in, speech-out conversation built on a multimodal model that does not transcribe to text in the middle. Sub-three-second latency, interruptible, with intonation and emotional nuance. Vision is integrated so you can point a camera at something while talking.
Key features
Sub-three-second response with natural interruption handlingLive camera input during voice conversationReal-time translation between languages, replacing standalone appsVoice style instructions, for example whisper, slower, more excitedSeamless switching between voice and text mid-thread
Best use cases
- Walking a site with the camera on and asking about defects in real time
- Live translating a conversation with a Polish or Romanian subcontractor
- Practising a pitch out loud and getting coached on pacing
- Hands-free dictation while driving or commuting
- Reading a French menu or a German tax form through the camera
Weaknesses
Will not sing or imitate named voices. Background noise still trips it more than the demos suggest. UK regional accents (Glaswegian, strong Scouse, Belfast) sometimes get mistranscribed.
Pricing
Short daily preview on Free. Materially higher caps on Plus and Pro. Available in app and web.
The closest thing to a hands-free intelligent assistant on a phone.
The long-form research mode that runs for up to 30 minutes, fans out web searches, reads sources, and writes a multi-thousand-word cited report. Lives inside ChatGPT now rather than as a separate product.
Key features
Up to 30-minute multi-source research runCited output with linked sourcesStrong on structured comparisons and market scansCan be combined with uploaded files for groundingNow backed by GPT-5.5 reasoning
Best use cases
- UK market sizing for a new service line with sources
- Competitor landscape sweep for a sales pitch
- Regulatory map for a sector you are entering
- Long-read explainer of a complex tax or compliance change
- Vendor due diligence pack from public filings and press
Weaknesses
Perplexity Deep Research is faster (two to four minutes) and arguably more honest about its citations. ChatGPT Deep Research sometimes pads with filler. Accuracy benchmarks put it behind Claude Code and Parallel Ultra on hard factual recall.
Pricing
Plus, Pro, Team, Enterprise. Limited monthly usage on Plus, generous on Pro.
Good enough when you are already in ChatGPT; Perplexity Deep Research is sharper.