Does it feel like the beginning of the end of ChatGPT?
A Redditor asked whether it “feels like the beginning of the end of ChatGPT”, saying they’ve cancelled their subscription and now use Gemini, Grok, Manus, Claude and Kimi for different reasons. You can read the thread here: Does it feel like the beginning of the end of ChatGPT?.
There are by far better models out there… ChatGPT is about keeping you on the platform, not giving the best answer.
It’s a fair question. In late 2025, no single model is “best” at everything. The landscape has shifted from a one-model world to a portfolio mindset: pick the right tool for the task. Below is a balanced look at where ChatGPT stands, how competitors compare, and what that means for UK developers and teams.
Why some users are moving away from ChatGPT
- Choice and specialisation: Models like Google Gemini, Anthropic Claude, xAI’s Grok and Moonshot’s Kimi have narrowed or surpassed ChatGPT on particular tasks (reasoning, coding, web-grounded answers, or longer context). This creates a strong case for multi-model workflows.
- Platform vs answer quality: Some users feel ChatGPT is optimised for platform engagement (custom GPTs, app-like experiences) rather than strictly the best one-shot answer.
- Release cadence: Competing labs have been shipping aggressive updates. Perception matters: if your day-to-day tasks feel faster or more accurate elsewhere, loyalty evaporates quickly.
Where ChatGPT still wins
- Ecosystem and tooling: ChatGPT’s “platform” approach can be an advantage if you want workflows without writing much code. For example, building a bespoke helper and pushing data into spreadsheets is straightforward. I’ve written a guide on how to connect ChatGPT and Google Sheets with a custom GPT.
- Safety and polish: OpenAI tends to ship conservative defaults and guardrails. For businesses, predictable behaviour can trump bleeding-edge capability.
- Breadth: Strong general performance across writing, ideation, light analysis and code assistance. Even where it’s not the absolute best, it’s rarely a bad choice.
How competitors stack up in late 2025
High-level impressions you’ll hear from practitioners (your mileage will vary by version and task):
- Google Gemini (docs): Often praised for image understanding, web-grounded answers via Google ecosystem, and tight integration into Workspace. Enterprise data controls are a selling point for regulated teams.
- Anthropic Claude (site): Favoured for long-context work (analysing big documents) and a cautious alignment approach. Many writers and analysts like its tone and reliability.
- xAI Grok (site): Strong on live web and social data if you’re in the X ecosystem. Researchers and power users like its up-to-the-minute feel for trending information.
- Kimi (Moonshot) (site): Notable for long-context capabilities and document-heavy tasks. Availability and language support may vary by region.
- Manus: The Redditor’s pick; not widely discussed in mainstream model roundups. If you’re considering it, check the vendor’s security posture, model card (if available), and UK legal terms.
Important note: headline benchmark scores are noisy. Real-world outcomes depend on prompt design, context window usage (how much text you can provide at once), tool calling, and retrieval-augmented generation (RAG – pulling in your own documents). Always test with your actual workloads.
UK-specific considerations: privacy, cost and compliance
- Data protection and DPAs: If you’re processing personal data, check whether the provider offers a Data Processing Agreement, data retention controls, and clear opt-out for training. UK GDPR expectations apply.
- Data residency and boundaries: Some vendors promise EU/UK data boundaries on enterprise plans; details vary and may not cover all features. Confirm scope in writing.
- Procurement and risk: Ask for audit logs, security certifications, breach notification terms, and model/version pinning. If you’re in finance, health or the public sector, involve your DPO early.
- Cost clarity: UI subscriptions hide token costs, but APIs don’t. If you move to programmatic use, build alerts and caps – costs can spike with large contexts or batch jobs.
- Availability and latency: Check regional availability and SLAs. For time-sensitive workloads, slow model responses can erase any quality gains.
When a multi-model strategy makes sense
- Research and web-grounded tasks: Try a model with strong live search/integration (e.g., Grok or Gemini) and verify sources.
- Long documents and policy analysis: Claude or Kimi are popular choices for extended context and careful reasoning on large files.
- Everyday drafting and ideation: ChatGPT remains a solid default if you value speed, polish and ecosystem features like custom GPTs.
- Team automations: If you need light workflow glue, ChatGPT’s platform approach (actions, integrations) can reduce engineering overhead.
How to decide: a quick evaluation checklist
- Define the task: Summarisation, code, data extraction, creative writing, policy analysis – different models shine in different places.
- Test with your data: Use a small, representative corpus. If you need RAG, include the retrieval step in your test.
- Measure what matters: Accuracy, latency, cost per output, and failure modes (hallucinations – confident but wrong statements).
- Check legal terms: UK GDPR compliance, data retention, training opt-out, and data location for your plan.
- Pilot, then standardise: Start with 1–2 models, write usage guidance for your team, and document known pitfalls.
So, is this the end for ChatGPT?
Unlikely. It’s the end of the one-model era. ChatGPT now competes in a healthy market where users will mix models to get the job done. Some will churn if they find better day-to-day answers elsewhere; others will stay for stability, safety and tooling. Both choices are rational.
If you’re in the UK and on the fence, run a short bake-off with your own tasks across ChatGPT, Gemini, Claude and one “live web” model like Grok. Score outputs, capture costs, and look at compliance terms. The “best” model is the one that meets your requirements with the least operational friction.
Useful links
