I keep getting asked which lab is ahead right now and what should be used for building things that actually have to work. Benchmarks are useful, but the daily experience of building is dominated by traits like reliability, UI capability, and whether the cost lets you iterate without second-guessing every prompt. This is my current read on where the major labs sit based on what I am reaching for when I need to ship something.
Anthropic
Anthropic is doing a great job, and it shows. Claude Opus is the model I hear the most consistent praise for from people who spend their time building. It is either the best or in the small handful of best models depending on what you are optimizing for. The biggest reason Opus keeps coming up for me is UI work. Most side projects are not math competitions; they are a pile of product decisions, UI glue, and front-end code that needs to look decent. Opus tends to be better at that whole bundle, which saves iterations.
Claude Code and Claude for Chrome are also strong products. If you are spending time inside an editor or browser and you want an agent that feels like it belongs there, that pairing makes sense. I wrote more about this in Claude Code + Opus 4.5: When the Model Finally Grows into the Harness. Claude Cowork is promising, but it is buggy. It is just where the tool is right now. If you are trying to depend on it for something time-sensitive, be ready to babysit it.
Subjective ranking of how well models handle the visual and product logic of UI design.
Google had a real comeback with Gemini 3, and usage is rising outside of the usual developer bubble. If you care about multimodal tasks, Google is ahead in a way that shows up quickly when you try to do anything involving images, mixed inputs, or complex media workflows. Nano Banana Pro is still the best image model in my view. If you are choosing an image generation default and you want the fewest annoying artifacts, this is the one I would bet on.
Gemini 3 Flash is also a standout for cost performance. If you need something that is good enough and you want to run a lot of calls, it is hard to ignore. I have a separate breakdown of the cost angle here: Gemini 3 Flash vs GPT-5.2 vs GPT-5 mini: Quality vs Cost in 2026. The gap for Google is reliability, especially around tool calling. When a model is unpredictable, it does not matter if it is smart on paper. Developers will keep a second model around as a safety net, and that tends to become the primary model over time. If the GA version of Gemini 3 Pro closes that reliability gap, Google is in a very good spot.
OpenAI
OpenAI is not my default for most things anymore. For the types of projects I spin up, UI ability matters a lot, and Opus is better for that. Where OpenAI still stands out is STEM. They do well on frontier math and that category of hard reasoning problems. That matters a lot to some teams, and it matters a lot less to others. For me, it is useful, but it is not the center of what I build.
ChatGPT is still useful for research and for memory. Memory is one of the most practical product features in this space, and OpenAI does it well. I also still have projects on OpenAI that I have not migrated, mostly because migrations have real switching costs even when they are easy to justify. The ChatGPT agent can be useful, but Claude for Chrome often replaces it for the workflows I care about. Code execution inside the web UI is also convenient when you want to do something quick without setting up a local agent.
On image generation, I put OpenAI behind Nano Banana Pro overall, but OpenAI occasionally wins on bizarre specifics. If you want more detail on the default differences you can spot between systems, I wrote about that here: ChatGPT vs Gemini Image Generation: Defaults, Artifacts, and Why You Can Tell Which Is Which.
xAI and Chinese Labs
xAI has cheap models, and that is the main reason to pay attention right now. The strategy seems to be smaller models that run faster, which lets them iterate quickly and keep pricing down. Grok 4 Fast and even Grok 3 Mini have been strong low cost options. I rarely reach for Grok 4.1 unless I am using the xAI web UI specifically for X search. Outside of price to performance, xAI is not leading for me. The interesting part is the rumored Grok 4.20 step up. If it closes the UI gap while keeping pricing low, that could change how often it shows up in developer stacks.
Z AI going public is a big signal. Their models are strong and cheap. GLM 4.7 is the budget version of Opus right now, and for a lot of coding tasks, it is in the same neighborhood as Sonnet 4.5, even if it takes more iterations. If cost matters a lot for your coding, GLM is an easy recommendation. One underrated angle here is throughput. If you use open models through providers that run them on high speed inference stacks like Cerebras or Groq, you can get extremely high tokens per second. That changes the experience more than most people expect, because waiting is a big part of iteration cost. Minimax M2.1 is another one worth watching as it behaves a lot like Opus and can be better than GLM 4.7 for coding in some cases.
ByteDance, Alibaba, Meta
ByteDance has Seedance 1.5 for video, and Seedream as a lower cost image model. It is not quite at the frontier image quality, but if you are generating a lot of images and cost is a constraint, it is a rational pick. Alibaba has Qwen and released Qwen 3 TTS. The audio model is good. The main place I see Qwen models show up usefully is in structured data tasks like editing JSON, where you want the model to behave more like a careful file editor than a chat buddy. Meta is not where I would send someone if they want the best experience right now. Maybe that changes later this year, but I am not planning around it.
If you want one primary model for coding and UI, I would start with Opus and only swap out when you hit a specific limitation. If you are building multimodal features or you care a lot about image generation, I would keep Gemini around and default to Nano Banana Pro for images. If you care about adoption trends and what normal users are doing, I have a separate post on chatbot market share here: AI Chatbot Market Share Jan 2026: ChatGPT Still Rules While Gemini Wakes Up.
