Which AI model is best? The answer depends on who is asking. For software engineers, the case favors Anthropic and its Claude models. For consumers, the case favors OpenAI and ChatGPT. Each company builds products around its models. The model produces the answers, and the software around the model is what a person opens and uses. The model is one piece of the choice. The software decides the rest.

Anthropic’s argument rests on Claude Code, a coding tool for developers. Claude Code works as an agent, a program built to plan its own steps and use tools to finish a task. Anthropic says the tool builds the plan, asks clarifying questions, and handles work running for hours or days. A developer hands Claude a bug fix, a test, or a multi-day migration, then directs and reviews the work from a terminal, an editor, Slack, or the web. The tool runs in the terminal and works alongside the editors and tools engineers already use, without a change to their workflow. Anthropic also gives developers the same tools and planning steps behind Claude Code, so developers build their own agents with them.

OpenAI’s argument sits with consumers, and the product around the model is ChatGPT. The service offers Free, Go, Plus, and Pro plans for individuals and plans for businesses. OpenAI says ChatGPT helps people learn, create, and get things done. The service covers three uses: chat for questions and everyday help, work for finishing tasks, and Codex for coding and technical work. The company’s headline describes one place to chat, work, and code. Codex works across ChatGPT, an editor, and the terminal, all connected by one ChatGPT account. The company says Codex completes tasks end to end, including building features and migrations. Coding is where the two companies’ products now meet.

Google also competes, with a release still due. The Gemini lineup lists 3.1 Pro as the Pro model, which Google describes as best for complex tasks, and marks 3.5 Pro as coming soon. The lineup also lists other models, including Gemini 3.8 Flash, 3.5 Flash-Lite, and 3.1 Deep Think. A new Pro model would show where Google stands.

Meta and xAI compete as well. Meta introduced Muse Spark in April, the first in a new series of models from its Superintelligence Labs, built for Meta’s products. At launch, the model powered the Meta AI app and website, with WhatsApp, Instagram, Facebook, Messenger, and AI glasses next. xAI, now part of SpaceX, builds the Grok models, and its newest, Grok 4.7, released September 21, targets coding and knowledge work.

China adds more companies. DeepSeek released V4-Pro on August 13 and V4.1-Flash on September 10, and lowered its prices for developers with the second release. The company also set off-peak prices at half the peak-hour price, effective August 16. Moonshot calls Kimi K3 its most capable model to date and says the model sustains long engineering sessions with minimal human oversight. ByteDance builds the Seed models, including Seed2.1, which ByteDance describes as a next-generation agent for real-world productivity. Alibaba builds the Qwen models, which Alibaba Cloud describes as open-source and multi-functional.

With this many models arriving, the products around each model separate the companies. The deciding factors are how deeply a company connects with businesses, the software around the model, and whether the tools help people finish work and build agents. ChatGPT sells plans to businesses, and Claude Code connects to Slack. OpenAI calls Codex the best way to build with agents. Google describes Gemini 3.8 Flash as best for complex agentic tasks, work carried out by agents. Meta says Meta AI launches several smaller agents at once to work on one question.

The case remains early. Companies keep releasing models, and the available materials describe no settled winner. The model is one piece of the choice, and the software around the model carries the rest. Attention now sits on how deeply each company connects with businesses and how well its tools help people build agents.

Sources

Anthropic, Claude Code by Anthropic | AI Coding Agent, Terminal, IDE — claude.com
Anthropic, Agent SDK overview — code.claude.com
OpenAI, Overview | ChatGPT — chatgpt.com
OpenAI, Codex | AI Coding Partner from OpenAI — openai.com
Google DeepMind, Gemini — deepmind.google
Meta, Introducing Muse Spark: MSL’s First Model, Purpose-Built to Prioritize People — about.fb.com
SpaceXAI, News: Research, Product & Company Updates — x.ai
DeepSeek, Change Log | DeepSeek API Docs — api-docs.deepseek.com
Moonshot AI, moonshotai/Kimi-K3 — huggingface.co
ByteDance, ByteDance Seed — seed.bytedance.com
Alibaba Cloud, Qwen — alibabacloud.com