There are thousands of AI chatbots out there, so how do you know which one to use? As an AI engineer in New Zealand, I’ve tested all the major ones and I’m going to rank them from worst to best so you can stop wasting time on the wrong one.
C Tier: Meta AI (Llama)
Meta’s Llama models are fully open source, which is great because you can run them locally on your own devices. However, in terms of benchmarks, they’re really far behind. The only area where they shine is throughput, meaning how fast the tokens come out. But for most applications, if the thinking and reasoning behind the scenes isn’t good enough, speed alone isn’t that useful. They have a Meta AI app and browser interface, but neither is particularly compelling compared to what else is available.
C Tier: DeepSeek
DeepSeek is fully open source and out of China. When it first released, it was pretty much on par with or better than the current GPT model at the time, and they trained it for a fraction of the cost. That sent shockwaves through the AI investment world and I remember share prices taking a hit. But unfortunately they haven’t been able to do much more since that initial release, so at this stage it drops to C tier. A year ago it would have been S or A tier.
A Tier (Microsoft) / B Tier (Personal): Copilot
If you’re in the Microsoft environment, Copilot is significantly better than average because it natively connects into Outlook, SharePoint, and all the information already in your ecosystem. But it’s really limited compared to other models. Behind the scenes it runs the ChatGPT 5 model, though there are different guardrails that minimise what you can actually do. Claude is becoming part of some Copilot applications too, but it’s still quite restricted. For business Microsoft users, A tier. For personal use, B tier.
A Tier: ChatGPT
The most well-known AI tool out there. Apparently 10% of the world’s population is using it weekly, which is a mind-blowing statistic. It’s been the real driver behind a lot of the AI innovation we’re seeing. The problem is that recently they’re falling behind. They used to sit at the top for benchmarks, but Google’s models and even some Chinese models are beating them now. It also feels like they’ve spread too thin with Codex, Atlas (their agentic browser), and other projects instead of focusing on the core model. You’re also much more locked into what they allow you to connect to. Other tools offer full customisability where you can connect to any MCP server or API and read and write freely, whereas ChatGPT controls that quite heavily. There’s a developer mode, but it doesn’t save conversations. I’ve cancelled my subscription recently, but it’s still very good for the average person.
A Tier: Perplexity
Perplexity is the best AI tool for research. The whole pitch is finding information online, whether that’s financial data, sports, analytics, or general research. It’s excellent at pulling back information, summarising it, and presenting it in a clean report format. Most importantly, it’s careful with referencing. A big early problem with AI was making up links to URLs, newsletters, and sources. Perplexity was the first tool that really solved that and made AI research reliable. If research is a core part of your workflow, this is the tool to use.
B Tier: Qwen
Qwen is from Alibaba. It’s fully open source like DeepSeek and it’s actually crushing some benchmarks, sitting in the top five for text encoding. Being open source means you can run it locally, and it has a mobile app and website. However, if you don’t speak Chinese, navigating the interface is a bit harder. Not impossible, but enough friction to keep it in B tier.
B Tier: Grok
Grok from Elon Musk’s xAI has probably been in the most controversies of any AI model out there, ranging from its outputs to what people are getting it to generate or say. That side is pretty rough. However, the benchmarks are decent. When the newest model drops it’s often sitting around the top five to ten. If you’re specifically looking to pull or extract information from X (Twitter), it’s probably the best option since it’s natively built on and trained on that data. Otherwise, I tend to avoid it.
B Tier: Kimi
Kimi is another Chinese open source model that you can run wherever you’d like. They recently released their K2.5 model and some benchmarks show it’s the best AI model currently out there, especially for coding and vision. But when models first release, it takes time to properly test and verify those results. For now it sits at B tier with potential to move up to A.
S Tier: Gemini
Google’s Gemini earns S tier just behind Claude. You get Gemini Flash for quick answers and Pro for deep reasoning. What I’ve found is it’s one of the best models for vision. If I upload a photo or give it a video as reference, it can interpret what’s going on and answer questions or help out accurately. Then there’s Imagen, which is the best image editing tool out there right now. I can upload a picture of myself, ask it to remove my glasses and put a red hat on, and it does it without distorting my face. Most other AI tools can put the hat on but then my face looks muddled. Imagen keeps it near perfect. They also have the Veo 3.1 model for video generation with native audio. I can create a video of a real estate agent walking through a house speaking throughout, and it sounds perfect while matching the visuals.
S Tier: Claude
My absolute favourite and my go-to for everything. There are a few reasons Claude sits at the top. First, Claude Code lets me develop websites, custom applications, and scripts really effectively. Second, the base Claude chatbot with MCP connections means I can connect Gmail, Notion, Xero for invoicing, and any other tool into one hub for my business and personal life. Third, Claude Skills work like custom GPTs but callable from one chat window. I can go from YouTube scripting to invoicing without context switching into separate agents. Even if you’re in the Microsoft or Google environment, I’d still recommend checking out Claude because it connects to all your software. Large language models are prediction engines. The easier they can connect into your business information, the better the outputs. Rather than copying and pasting information in and out, direct connections make everything faster and more accurate.
The Full Tier List
S Tier: Claude, Gemini. A Tier: ChatGPT, Perplexity, Copilot (Microsoft users). B Tier: Qwen, Grok, Kimi, Copilot (personal). C Tier: Meta AI (Llama), DeepSeek.
That’s my chatbot tier list going into 2026. The landscape changes fast, but the fundamentals stay the same: pick the tool that connects best to your existing systems, don’t chase hype, and focus on what actually delivers results for your workflow.
