Claude vs ChatGPT vs Gemini vs Grok (September 2026)

• Best for coding, agents and serious writing: Claude. Opus 5.5 leads GPT-6 Astra on most of the shared benchmarks Anthropic published, and Claude Pro is $20/month.
• Best all-rounder: ChatGPT. GPT-6-family models for paying users (sources disagree on which one each plan gets in chat), plus voice, images and the widest app ecosystem.
• Best for Google Workspace users: Gemini. Gemini 3.1 Pro remains the live Pro model; the next one is late.
• Best for real-time X data and low API prices: Grok. Grok 4.7 launched 21 September at $2/$6 per million API tokens.

• Claude: Sonnet 5 arrived on 30 June, Fable 5.1 on 1 September, and Claude Opus 5.5 became the flagship on 22 September. Opus 5.5 raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans and added a bankable rate-limit reset. Plan prices did not change.
• ChatGPT: GPT-6 Astra reached general availability on 4 September. GPT-6 Sol and GPT-6 Luna followed on 22 September, about 90 minutes after Opus 5.5 (TechCrunch). OpenAI also loaded a banked reset into Plus, Pro and Business accounts that day.
• Gemini: Gemini 3.8 Flash is Google's current Flash model. Gemini 3.5 Pro was announced at I/O on 19 May and has since missed June, July and August targets. As of 28 September, Google DeepMind's Koray Kavukcuoglu says Gemini 4 is in early post-training, with no release date.
• Grok: xAI released Grok 4.7 on 21 September.

USD first, approximate GBP in brackets at £0.79 per dollar, excluding tax. Claude's prices are confirmed for this update. The ChatGPT, Google and Grok prices are the figures from our June 2026 check. We have no verified source for them after the September launches, so confirm them on each vendor's site before you pay.

The entry tiers are still within $10 of each other. The bigger September change is what $20 buys. Anthropic raised Claude's five-hour usage limits with the Opus 5.5 launch, and both Anthropic and OpenAI added a banked rate-limit reset you can save for a heavy day. For the full Claude breakdown, see our Claude pricing guide and Team vs Max comparison.

No single benchmark covers all four assistants. The cleanest head-to-head is Claude against ChatGPT. Anthropic's Opus 5.5 launch table puts Opus 5.5 at 66.4% on Terminal-Bench 4.0 against 57.9% for GPT-6 Astra, and at 1846 Elo on GDPval-AA v2.1 against 1542. Astra leads on AutomationBench (41.4% vs 40.0%) and Terminal-Bench-Science (64.6% vs 58.7%). The Astra figures are as reported by OpenAI. The Claude figures use max effort with safeguards on.

Google and xAI are missing from those tables. The best cross-vendor signal we have is the independent Artificial Analysis Intelligence Index. As of 29 September 2026 it scores Opus 5.5 and Sonnet 5.5, but has no score for any Gemini model in our sources, so the chart below shows only the models it does cover. The one Gemini data point we have is GPQA Diamond, as reported by OpenAI: Gemini 3.8 Flash scores 95.3 against 96.0 for GPT-6 Astra and 93.7 for Claude Fable 5.1. That is a strong result for Google's cheap model.

Claude Opus 5.5 now tops the index at 58, with Sonnet 5.5 at 56, ahead of GPT-6 Astra and Fable 5.1 on 53 and GPT-6.1 Sol on 52. Grok 4.7 trailed at 46 at our 23 September check. For every number across all the September launches, see our frontier AI benchmarks tracker and the Opus 5.5 vs GPT-6 Astra head-to-head.

If you are building on these models rather than chatting with them, the gaps are much wider than on subscriptions. Standard-tier list prices, USD per million tokens:

Per-token price is not cost per task. Models use different numbers of tokens for the same job. Anthropic says Opus 5.5 uses fewer tokens per task than Opus 5, and independent testers describe Grok 4.7 as slow and verbose. On our own list-price model of a coding agent (50M input tokens a month at 90% cache hits, 5M output), Opus 5.5 costs $129.00 a month against $345.00 for GPT-6 Astra, $160.00 for Gemini 3.1 Pro (no cache discount assumed) and $62.50 for Grok 4.7.

Two mid-tier models share a headline price. GPT-6.1 Sol and Claude Sonnet 5.5 both cost $2 input and $10 output; Sol's cached input is $0.10 against Sonnet 5.5's $0.20. The full workload model and every tier are in our AI API pricing comparison, and Claude's rate card is in Claude API pricing.

If you pick Claude, these are the jobs where it pays for itself.

1. Ship real software with Claude Code

Claude Code is the agentic coding tool that runs in your terminal and editor. It reads a whole codebase, plans multi-file changes, runs tests and fixes its own errors. Anthropic reports that one early tester completed a 680,000-line migration with Opus 5.5 in under a day. In another test, Opus 5.5 succeeded 39 times out of 40 at speeding up a web app.

2. Write and edit long, technical documents

Reports, documentation, proposals and essays. Claude's prose is precise, and it tells you when it is unsure instead of bluffing. That matters for anything you will put your name to.

3. Research and analyse documents

Give Claude PDFs, contracts, spreadsheets, screenshots or diagrams, and ask it to summarise, extract, compare or critique. With 1M tokens of context at standard pricing, very large document sets fit in one session.

4. Build agents and automations

Through the Claude API and the Model Context Protocol (MCP), Claude connects to GitHub, Slack, databases and your own systems, then acts on your behalf. See our connectors directory for what it can plug into. Claude Cowork extends this to your desktop.

5. Day-to-day knowledge work

Drafting emails, planning, brainstorming, learning a topic and turning messy notes into something usable. For most people, Claude Pro at $20 (approx £15.80) a month covers all of this, and the September limit increase gives you more room.

• Not Claude if you need image generation or a polished voice mode. Also skip it for offensive security work: most cybersecurity tasks on Opus 5.5 are re-routed to Opus 4.8 unless you join Anthropic's Cyber Verification Program. Thinking cannot be switched off on Opus 5.5. On business-process automation (AutomationBench) and scientific research (Terminal-Bench-Science), GPT-6 Astra scores higher.
• Not ChatGPT if you build heavily on the flagship API. At $10/$50, Astra costs 2.5 times Opus 5.5 per token. Astra reaches "Critical" cyber capability under OpenAI's Preparedness Framework and refuses to write proof-of-concept exploits. The UK AI Security Institute found it could evade monitoring under adversarial prompting. Free and Go users are reported to get GPT-6 Luna, OpenAI's cheapest model, rather than Astra, though sources disagree on the exact per-plan mapping.
• Not Gemini if you want the newest frontier model today. Gemini 3.5 Pro has missed three release targets, so you are on Gemini 3.1 Pro, and it does not appear in Anthropic's or OpenAI's head-to-head tables. Gemini 3.8 Flash's API price doubles after 31 December 2026.
• Not Grok if cost per task or peak reasoning matters most. Grok 4.7 scores 46 on the Artificial Analysis index against 53 for Astra and Fable 5.1, testers find it slow and verbose, and SuperGrok had the highest entry price of the four at our June check.

• Choose Claude if you code, build agents, or write seriously and value reliability over extras.
• Choose ChatGPT if you want one tool that does everything well, including the GPT-Live voice mode, images and the largest ecosystem.
• Choose Gemini if you live in Google Workspace, or want a cheap, fast model through the API (Gemini 3.8 Flash).
• Choose Grok if real-time X data and a less filtered style matter to you.
• Run two if you can. Many professionals pair Claude for building and writing with ChatGPT or Gemini for multimodal work and search. At about $20 a month each, a two-tool setup is affordable.

September reset the field. Claude Opus 5.5 is now the strongest pick for coding and professional work on the shared benchmarks, and it costs well under half of GPT-6 Astra through the API. ChatGPT is still the safest all-rounder, with GPT-6-family models on every paid plan (which one you get in chat varies by report, so check your model picker). Gemini is waiting on a late 3.5 Pro. Grok 4.7 is the cheapest flagship per token, though testers find it slow and verbose, which pushes up cost per task.

Subscription prices are close, so choose on the work. If you have decided on Claude, our pricing guide shows which plan fits. If you are comparing OpenAI's new tiers, start with GPT-6 Sol vs Astra vs Luna. For background, our June 2026 model wars roundup covers the previous generation, and the 100,000-human benchmark compares these models against people.

• Anthropic: Claude Opus 5.5 announcement
• Anthropic: Claude API pricing docs
• OpenAI: GPT-6 Astra
• OpenAI: Introducing GPT-6 Sol and Luna
• VentureBeat: GPT-6 Sol and Luna pricing
• TechCrunch: OpenAI launches GPT-6 Sol and Luna
• Vellum: GPT-6 Sol and Luna benchmarks explained
• DataCamp: GPT-6 Astra
• Artificial Analysis: GPT-6 Sol