In a single week, five major players shipped a coding or agent model. Google introduced Gemini 3.7 Flash, built for coding and agents. xAI made its Grok 4.6 coding model available inside GitHub Copilot, and Elon Musk teased that the incoming 4.7 will be more capable. DeepSeek pushed V4-Pro to general availability and released an open-source tool called Harness. Writer launched Palmyra X6 with a rebuilt agent harness for enterprises. And Z.ai shipped GLM-5.3.
This is a race, but right now, it seems the winner isn’t chosen yet.
Five models, one week
The timing is the signal. When Google, xAI, DeepSeek, Writer, and Z.ai all ship coding models in the same week, the race to own the AI coder has officially become the most crowded contest in the industry. Every major lab is betting that the developer is the highest-value user, because whoever controls the tool that writes software actually shapes the digital world.
The AI coding race splits in two
The interesting part is the split. The race is heating up, yes, but the tactics, even the long-term strategies, look quite different. On one side are the closed frontier labs, Google with Gemini and xAI with Grok, pushing their models into the tools developers already use, like GitHub Copilot. On the other side are the open-source upstarts. DeepSeek with Harness and Z.ai with GLM-5.3 literally giving the code away and competing on cost and openness.
The DeepSeek lesson applies here. A model can top a benchmark and still stumble on real tasks, so the race isn’t only about raw capability, simply because there are many fields to compete on.
The most important question is which setup is reliable for which task, and which one developers trust with their code.
What this means for developers
For developers, the short version is that choice is multiplying and the price of AI coding is under pressure. Which is good, because from the viewpoint of the frontier labs, developers are end-users, just like the average person with a monthly subscription, and competition is good for the market. Lower prices, better products, wider selection of features. The end-users win. Open-source models push costs down, while closed labs bundle theirs into the tools already on your screen.
The real test is not the launch, because there are launches every few months now, with top models. A race where you can’t let yourself slow down, because if you do, the competitors will catch you. So what matters? How can a lab say our stuff is good?
The benchmark of the pudding, that’s how. If these models hold up on real codebases day after day, then the labs can say “we’re good”.
For everyone else, this matters due to a really simple fact. Software is infrastructure. The models that win will shape the tools that build the apps and systems we rely on. And not just us, but the whole market, consumers and producers, fabs and service providers, everyone. Whoever owns the AI coder will have a lot of say over what gets built.
So the arms race just got five new weapons. And the winner is still not decided yet.
