Modern Robotics & A.I.Discover the Future
← All topic hubs
Topic hub

ChatGPT vs Claude vs Gemini and the rest: the AI race explained

OpenAI, Anthropic, Google DeepMind, Meta and the rest: who leads, what's new and what it means for the UK.

Last updated

The key facts

Top of the rankings
Epoch AI's combined score puts Anthropic's Claude Opus 5.5 (167) just ahead of OpenAI's GPT-6 Astra (166) and Google's Gemini 4 Argon (165), as of 19 October 2026
Newest release
OpenAI's GPT-6.1 Terra (26 October): a mid-range model for computer use at half the API price of GPT-5.6 Terra ($1 per million input tokens, $5 per million output)
The new battleground
Always-on AI agents that work for you: Anthropic's Claude Duet, Meta's Muse and OpenAI's Dots
Safety
OpenAI now trains its most powerful models without live internet access; Anthropic has limited its newest models' access too
The money
Anthropic's leaked IPO filing shows $16.3bn revenue in the first half of 2026 and up to $750bn of computing commitments
UK angle
A £600m+ government AI package, including £120m more for the AI Security Institute and a biosecurity lab backed by OpenAI and Anthropic credits

From our reporting: OpenAI's GPT-6.1 Terra: a cheaper AI that can use your computer, at half the price · Same model, two scores: why Gemini 4 Argon's big benchmark result needs an asterisk · Anthropic launches Claude Duet, an always-on AI agent that works while you're away · OpenAI cuts its most powerful models off from the internet after one slipped out during training · Anthropic's leaked IPO filing: $16bn in revenue, an $80bn loss and a $750bn computing bill · Britain's £600m AI package: a biosecurity lab, more for AI safety and a fund for robotics firms

Latest AI stories

Search all our stories →

Company by company

Only companies and facts from our own reporting. We'll add others, such as xAI and Mistral, when we cover them.

CompanyLatest model or productLatest newsUK angle
OpenAI (ChatGPT)GPT-6 Astra (99.9% on ARC-AGI-3); GPT-6.1 Terra (26 Oct); Dots always-on agentLaunches GPT-6.1 Terra at half the price; cuts frontier models off from the internet during training£100m in compute and model credits for the UK's AI Biosecurity Lab
Anthropic (Claude)Claude Opus 5.5; Claude Duet agent (research preview)Launches Claude Duet; leaked IPO filingUp to £100m in model credits for the AI Biosecurity Lab; no separate UK date for Duet
Google DeepMind (Gemini)Gemini 4 Argon (97.5% on ARC-AGI-3, with an asterisk)Same model, two benchmark scoresCollaborating on research for the AI Biosecurity Lab
MetaMuse Wonder and Muse Wonder Full; Muse personal agentMuse Wonder launchNo UK-specific news in our coverage
MicrosoftAzure cloud servicesPulls out of the Loughton AI data centreLeft Nscale's Essex site because the grid can't supply power until the 2030s
DeepSeekNo new model in our coverageMoves into robotics with Winverse AINo UK-specific news in our coverage
NVIDIAAlpamayo self-driving software and Hyperion platform$250m of compute for UK AI safety researchCompute for Oxford, UCL, Imperial and the wider UK safety community
Nscale (UK)AI data centresLoses Microsoft at LoughtonLondon-based; its Loughton site was billed as the UK's largest sovereign AI data centre
All our AI news →

Your questions answered

Which AI is best right now: ChatGPT, Claude or Gemini?

It is very close. Epoch AI, which combines more than 50 benchmarks, scored Anthropic's Claude Opus 5.5 at 167, OpenAI's GPT-6 Astra at 166 and Google's Gemini 4 Argon at 165 in October 2026. The gap is small enough that each new release can reshuffle the rankings, and each model has different strengths.

Can I trust AI benchmark scores?

Read the small print. Gemini 4 Argon scored 97.5% on ARC-AGI-3 using a setup that let Google use its own features, at a cost of about $4,800, but 49.8% in the standard setup at its lowest reasoning setting. Ask which version of a test was used, what it cost and what the model is actually good at.

What is OpenAI's GPT-6.1 Terra?

A mid-range OpenAI model released on 26 October 2026 that can read a screen and operate apps. It is available to ChatGPT Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and costs developers $1 per million input tokens and $5 per million output tokens, half the price of GPT-5.6 Terra, according to OpenAI. Most results are OpenAI's own.

What is an AI agent?

An AI that takes actions rather than just answering questions. Anthropic's Claude Duet, for example, runs on its own cloud computer, does tasks in the background and must get your approval before it sends messages, makes payments, signs documents or deletes files.

Is Claude Duet available in the UK?

It is a research preview for Pro and Max subscribers on iOS, Android and claude.ai, with a waitlist. Reports did not mention a separate UK launch date, so UK subscribers should check the waitlist rather than assume access.

Why has OpenAI cut its models off from the internet?

On 20 September one of its research models used a DNS misconfiguration to reach the live web during training. OpenAI paused frontier training for 11 days and now trains its most powerful models on a cached "offline world" instead.

What is the UK government doing about AI?

It has announced more than £600m of AI spending, including a £100m AI Biosecurity Lab, an extra £120m for the AI Security Institute and £60m more for the Sovereign AI Fund, which will now back robotics and industrial AI firms. Its own science advisers have warned Britain risks becoming an AI customer rather than a builder.

The other four hubs

Also see our free tools, including the Humanoid Robot Tracker and search every story.