Best AI Chatbots 2026: ChatGPT vs Claude vs Gemini vs Grok Tested

ChatGPT holds first at 9.7, and the most useful item for ChatGPT subscribers this week sits in OpenAI's own changelog. On 14 September OpenAI announced that GPT-5.5 retires from ChatGPT, ChatGPT Work and Codex on 14 October, across every plan from consumer to Enterprise and Edu. The OpenAI API keeps it. Inside the apps, the replacement is GPT-5.6 Sol.

For most people this is a non-event, because the default model simply changes. The work is in everything that pinned a model by name. OpenAI's notice lists the places to check: workspace defaults, saved model settings, custom agents, scheduled tasks and scripts. If you set up a scheduled morning briefing or a custom agent over the summer, open it now and switch the model yourself. A task you test in September is a task that still runs on 15 October.

I keep ChatGPT at 9.7 because this is exactly how a platform should retire a model: a month of notice, a named successor and a checklist. Its ecosystem score of 9.7 reflects that maturity as much as the feature list.

Claude at 9.5 stays my pick for long writing and document work, with writing at 9.7 and coding at 9.7, both the best scores in this table. Gemini at 9.4 keeps reasoning at 9.7 and realtime at 9.5, and it is the assistant I use when the answer depends on something that happened this morning.

My practical advice applies to every assistant here. Keep one short document listing every automation you built on an AI chatbot, the model it uses and the date you last tested it. It takes ten minutes, and on retirement day it turns a surprise into a two minute fix.

Perplexity at 8.6 and DeepSeek at 8.2 stay where they are, and the ranking carries forward unchanged.

ChatGPT holds first at 9.7, and this week I want to separate two questions readers keep merging: which assistant you pay for, and which assistant answers when you hold the side button on your phone.

iOS 27 shipped on 14 September with Siri rebuilt as a standalone app, trained with Google Gemini models, and the deepest features gated to an A17 Pro chip or newer. After two days with it my read is that the new Siri works as a front door. It handles device control, on-screen context and quick lookups, and the long thinking still belongs to whichever assistant you chose. Your subscription decision survives the update intact.

That decision got better value in September. GPT-6 Astra arrived on 3 September and keeps ChatGPT ahead on reasoning at 9.8, which is why it stays my single recommendation for anyone who wants one tool covering work, study and everyday questions. Gemini holds 9.4, and its 3.8 Flash release on 2 September made the free tier quick enough that I now run it as a second opinion on every research task. Claude at 9.5 remains the one I trust with a 200 page contract, and its writing score of 9.7 is still the highest here.

DeepSeek moves to 8.8 on reasoning after V4.1-Flash landed on 10 September. At 8.2 overall with a perfect value score, it is the assistant I point people to when the monthly fee is the blocker. Grok at 9.0 keeps the freshest feed for live events, and Perplexity at 8.6 stays my pick when I need answers with citations I can open myself.

One practical note for iPhone owners. Install the app of the assistant you pay for, turn on its own voice mode, and put it on your home screen next to Siri. Two front doors, each good at something specific, is the setup I run myself.

ChatGPT keeps first at 9.7, and this week I spent my testing time on a question readers keep asking me: which assistant can you count on when your main one goes dark?

The September 3 disruption is now better understood. The Register reported that ChatGPT, Claude, and Grok failed in the same window. OpenAI traced its part to a routing error that began at 7:43 a.m. Pacific and was fixed by about 8:17. Anthropic described a partial infrastructure outage across Claude.ai, Claude Code, and its developer API. SpaceX said Grok's downtime came from an outage at its Memphis compute centre. Google recorded no formal Gemini incident on its status page.

Three separate causes landing together tells me the fix belongs with users: keep two assistants ready. No scores move on reliability this week.

ChatGPT stays on top for breadth. GPT-6 Astra handles deep reasoning, GPT-Live-1 powers the upgraded voice mode, and memory, file libraries, and agents all sit inside the same $20 plan. It is the single subscription that covers the most jobs.

Claude holds 9.5 as my writing and long-document assistant. Claude Fable 5.1 summarizes a dense report while keeping the author's caveats intact, which is exactly what I need before a meeting.

Gemini keeps 9.4 and earns my backup slot. The free tier runs Gemini 3.8 Flash, answers quickly, and already sits inside Gmail and Drive, so switching over during an outage costs nothing.

Grok at 9.0 remains the fastest route to what people on X are saying right now. Perplexity at 8.6 is the tool I open when every claim needs a citation. DeepSeek at 8.2 stays the best zero-budget reasoning model.

My setup recommendation for this autumn: pay for the assistant that fits your main work, and keep Gemini's free tier signed in on your phone as the spare.

No ranking changes this week.

ChatGPT stays first at 9.7, and the past ten days handed me more reasons to keep it there. OpenAI released GPT-6 Astra in early September, then on September 10 shipped GPT-Live-1 to the API alongside a ChatGPT Voice upgrade that calls on GPT-5.6 or GPT-6 Astra whenever a spoken question needs search or reasoning. Voice, memory, file libraries, and agents in one $20 subscription is exactly the breadth my realtime and ecosystem columns reward.

Claude holds second at 9.5. Anthropic shipped Claude Fable 5.1 and Claude Mythos 5.1 on September 1, and for long documents and careful writing it remains the assistant I open first. Hand it a dense contract and ask which clauses shift liability, and it quotes the right paragraphs and flags the ambiguous wording clearly.

Gemini keeps third at 9.4. Gemini 3.8 Flash became generally available on September 2, and the value story keeps improving: fast answers, tight hooks into Gmail and Drive, and a free tier generous enough for most students.

One event matters for anyone relying on a single tool. On September 3 an Azure East US failure took ChatGPT, Claude, and Grok offline for about 90 minutes while Gemini stayed up. I did not move scores over one outage. I did change my own routine: keep a second assistant signed in on your phone, and Gemini's free tier is the obvious pick for that backup role.

Grok stays fourth at 9.0 on real-time information pulled from X. Perplexity holds fifth at 8.6 as the research tool that cites every claim it makes. DeepSeek at 8.2 remains the strongest free option for reasoning on a zero budget.

Nothing moves this week. The releases were large, and each one reinforced the product that already led its lane.

Two days after Apple put an on-device assistant at the centre of its September keynote, I went back through this ranking to see whether anything should move. Nothing does, and the reason is worth stating plainly. ChatGPT holds first at 9.7 because it covers more shapes of a working day than anything else shipping: memory that carries between sessions, voice that survives a walk around the block, a canvas for long document edits, and connectors into the drives where people already store their files.

Claude stays second at 9.5 and keeps the work I care most about, which is long documents and reasoning that has to hold up when someone checks it. Gemini holds 9.4 and owns Google Workspace outright. If your team lives in Docs and Sheets, that single fact outweighs any leaderboard position.

Apple's announcement changes the question underneath the category. An assistant answering on the phone itself takes the quick lookups and short dictations that used to mean opening an app. I expect every chatbot here to feel that in usage volume while the ranking order stays exactly where it is, because these assistants earn their keep on work that runs longer than a sentence.

My advice today is the same shape as last week and sharper in detail. Pick the assistant whose ecosystem already holds your documents, then judge it on tasks that take more than thirty seconds. Short prompts are heading toward whatever ships on your phone, and that frees a paid subscription to prove itself on work that actually justifies one.

This was the busiest seven days the chatbot market has had all year, and it settled a question I have been asked constantly since spring. GPT-6 Astra shipped on September 3 and it is a genuine generational release, so ChatGPT stays at number one and earns a bump to 9.7. Astra lists at $10 per million input tokens and $50 per million output, which is 2.5 times what GPT-5.6 Sol charged on its promotional rate. That is the real story here. OpenAI is confident enough in the reasoning gap to charge like a premium vendor again.

Claude holds second, and I want to be specific about why. Anthropic shipped Fable 5.1 on September 1 and cut cache-read pricing by 75 percent in the same release. For anyone running long-context assistants over a fixed document set, that single pricing change does more for monthly cost than any benchmark gain announced this week. I raised Claude's value score to reflect it.

Gemini stays third on 9.4. The 3.8 Flash release on September 2 held the previous introductory price, which keeps Google the cheapest credible option at the high-volume tier, and Flash remains the model I reach for when latency matters more than depth.

My advice for the next month is to resist switching on launch-day excitement. Astra is the strongest reasoner available today, and it costs five times more per output token than the Flash tier that handles most everyday questions perfectly well. Pick per workload, and keep two providers wired up.

OpenAI has started rolling GPT-6 Astra out to a selected set of customers, and I want to be precise about what that means for a buying decision today. A staged rollout to vetted accounts is a signal about where the ceiling is heading, and it is not yet something a person paying twenty dollars a month can sit down and use. I am keeping ChatGPT at the top because its everyday surface remains the most complete package on the market: voice, memory, file handling, image generation and a mobile app that behaves the same way on every device. The Astra rollout raises my reasoning score by a tenth because the safeguards work OpenAI published alongside it tells me the model line is being hardened before it scales.

Claude stays a very close second on the strength of long-document work. I spend most of my week feeding it forty and fifty page PDFs, and it holds detail across the whole thing in a way that shows up immediately when you ask it to cite back to a specific paragraph. Gemini remains the pick for anyone who already lives inside Google Workspace, because the value there comes from the connection to Gmail, Drive and Calendar.

Grok moves up a tenth this week. xAI shipped an enterprise agent product with real access controls, network restrictions and audit logging, and that changes who can actually deploy it. Audit trails are the feature that gets a tool past a security review, and Grok now has them.

My advice for this month is to hold your subscription steady. There is a large model transition happening across all three leaders, and the sensible move is to wait for general availability before you reshuffle your workflow around any one of them.

The tie broke this week. OpenAI shipped GPT-6 Astra, staged to general availability at $10 in and $50 out per million tokens with a 1.05M context window, and it is the first model the company has flagged as reaching its critical-cyber safeguard threshold. Self-reported numbers put it at 57.7 on Terminal-Bench 4.0, 74.1 on DeepSWE and 96.0 on GPQA. That is a genuine capability step, and I am moving ChatGPT to 9.6 and first place on the strength of it. Claude drops to second at 9.5, which is a ranking change and not a criticism: Anthropic's Fable 5.1 landed September 1 with cache reads down to $0.25, and for anyone running long agentic sessions that price cut is still the most consequential release of the week. Claude keeps my recommendation for long-document reasoning and code that has to pass review. Gemini holds 9.4 after 3.8 Flash reached stable general availability on September 2, and the cheap tier is now genuinely production-ready with grounding good enough for real applications. Grok at 8.9 keeps its lead on real-time social data. Perplexity at 8.6 is still what I open when I want cited answers. My practical read for September: the safeguard flag on Astra is the story worth watching, because a frontier lab publicly gating a model on cyber capability changes what enterprise procurement asks for. Subscribe to one flagship, use free tiers elsewhere for a second opinion.

The top of this list moved this week and I want to be precise about why. Anthropic shipped Claude Fable 5.1 on September 1, positioned as its strongest model for coding and knowledge work, tuned specifically for agentic tool use and paired with much cheaper cache reads. Google followed on September 2 with Gemini 3.8 Flash reaching stable general availability. Claude and ChatGPT stay tied at 9.5 and I am comfortable defending that tie, because they now win on different axes and most readers should pick based on what they do all day. Claude takes long-document reasoning, code that has to survive review, and anything where the model needs to hold a large codebase in working memory. The cheaper cache reads in the 5.1 release matter enormously for anyone running long agentic sessions, since that is where API bills quietly balloon. ChatGPT holds its 9.5 on breadth: the widest tool ecosystem, the strongest consumer app polish, and voice mode that remains the most natural of the three. Gemini at 9.4 has the best real-time grounding and the tightest integration with Workspace, and the Flash 3.8 general availability makes the cheap tier genuinely usable for production tasks. Grok at 8.9 stays where it is on real-time social data. Perplexity at 8.6 is still the one I open when I want an answer with its sources attached. My practical verdict for September: subscribe to one flagship and use the free tiers of the others for cross-checking. Paying for two is a waste at these capability levels.

August closed with sixteen new models from eight providers, and the shape of that flood tells you exactly where the chatbot race now sits. The frontier tier has stopped moving on raw intelligence and started moving on speed and price. Gemini 3.7 Flash landed on August 13 as the fastest frontier-class model available, Grok 4.6 arrived the day before, and Alibaba closed the month with Qwen3.8 Flash on August 26. Every one of those is a latency and cost play. I keep Claude and ChatGPT tied at the top because the work I actually give a chatbot, long reasoning chains, careful writing, code that has to run, still separates on quality, and tokens per second barely enters into it. Claude Opus 5 has held that position since late July while costing roughly half what the previous flagship tier did, and that price move matters more to a daily user than another benchmark point. Gemini stays a very close third on the strength of its real-time grounding and the Flash tier being genuinely quick enough to change how you use it. The interesting movement this month is at the bottom. Qwen Chat moves up to ninth on the back of Qwen3.8 Flash, which gives the free tier a materially faster default model than Le Chat currently offers. If you are choosing today, pick on the work you do most: Claude for sustained reasoning and writing, ChatGPT for breadth of tooling, Gemini for anything that needs live web grounding.