Who's Winning? Five Frontrunners, Five Different Races

For the past few years, Forward Future and the rest of the AI industry have been obsessed with one question: Who has the best model? Seriously, we talk about it every week.

In 2023, GPT-4 was the benchmark everyone else was chasing. OpenAI established the standard against which competing models were measured. Then Anthropic's Claude 3.5 Sonnet arrived in 2024, establishing Claude as a serious contender for developers. Google closed the gap. Now SpaceXAI and Meta are making an aggressive push into frontier intelligence.

But these companies aren't just competing to build the most capable models. They're pursuing very different paths to winning the AI market.

The company with the smartest model won't necessarily capture the most value. And the company that wins over consumers might not be the one that dominates enterprise workflows, autonomous agents, or infrastructure.

Here's my take: We're treating AI like a single race when it's increasingly becoming several overlapping competitions, each with different advantages, economics, and definitions of success.

The next phase of the AI race will be decided by who controls the customer relationship — which we discussed here — who can deliver intelligence at the right price and who makes AI indispensable to everyday work.

Every advance in model intelligence is incredibly exciting, and it will always be important. Strong reasoning, reliable coding and accurate execution are prerequisites for useful AI.

For many everyday use cases, frontier models are now capable enough that five other factors can matter more than another incremental gain in raw intelligence:

A brilliant model that forgets your goals, can't access your tools or costs too much to use has limited practical value for the majority of users. But a slightly less brilliant model that is integrated with all your work tools and that your company can afford to provide to everyone? That will be the model people rely on.

Take something as simple as following up with a potential Forward Future sponsor. If my AI remembers the pricing we discussed, the placements we considered, and where we left off, it can help draft a relevant response immediately. I don't particularly care whether it's the most powerful model if it saves me from digging through old emails or explaining everything again.

That kind of utility can shape which products people adopt and, more importantly, which ones they keep using.

And each frontrunner is approaching that opportunity from a different starting point.

OpenAI has an advantage none of its competitors can easily replicate: ChatGPT helped establish the habit of turning to AI for answers, ideas and work. And it’s still the most popular.

That direct relationship with individual users is enormously valuable. Millions of people already know where to go when they need AI, and OpenAI is expanding what they can accomplish once they get there.

Its strategy extends beyond frontier models like GPT-6 Astra. Through ChatGPT, developer tools, enterprise offerings and increasingly capable agents, OpenAI is building an ecosystem designed to keep users coming back for more of their work.

At $10 per million input tokens and $50 per million output tokens, GPT-6 Astra commands a substantial premium over its rivals, putting greater pressure on OpenAI to demonstrate that its performance justifies the cost.

The bigger prize is becoming the primary interface between people and AI, regardless of which underlying model or application ultimately performs a task.

That position could give OpenAI enormous influence over how users discover tools, delegate work, and interact with software.

But there's a vulnerability in that strategy: The AI interface doesn't necessarily have to be a standalone destination.

Google can bring AI directly into Gmail, Chrome, and Android. Microsoft can embed it into workplace software. Meta can make AI available inside the messaging applications people use constantly.

If AI becomes a capability built into everything rather than a destination people visit, OpenAI's existing ChatGPT habit may become less defensible.

OpenAI is betting that the company people learned to turn to first can remain the primary place they go to get things done.

Google's greatest competitive advantage is that it already owns critical parts of how billions of people find information, communicate and get work done via search, Gmail, Drive, Chrome, Android and Google Cloud.

Rather than persuading everyone to adopt a new AI product, Google can bring increasingly capable models directly into existing workflows.

At the same time, its new Gemini 4 Argon demonstrates that it remains a formidable frontier-model competitor.

Designed for complex professional work, Argon targets coding, finance, law, cybersecurity and scientific research. Google reports that teams using it through Antigravity increased agentic code submissions by more than 35% while reducing rollbacks by nearly one-third.

Those are company-reported results, but they point toward a performance measure that’s attractive to many business buyers: work completed.

Google is also competing aggressively on price, offering Argon at an introductory $2 per million input tokens and $10 per million output tokens before those rates double in December.

The combination of a comprehensive, widely used suite of tools with an excellent model is especially powerful in the enterprise, where companies have strong incentives to consolidate vendors to control costs, manage risk and reduce operational complexity.

Google's objective is to make AI the default experience within the software people already use.

But distribution doesn't guarantee adoption.

People already have access to Google products every day, yet that hasn't prevented ChatGPT or Claude from building substantial followings. Users will still seek out specialized products when they offer a meaningfully better experience.

And there's another tension worth watching. Google has an enormous existing business built around how people search for information and interact with software. AI agents could fundamentally change both behaviors, forcing Google to balance protecting existing revenue with embracing new models of interaction.

Here's where the competition gets especially interesting. Google and OpenAI are competing to build frontier models, yet OpenAI has announced that Google Cloud will offer its models, including GPT-6 Astra.

OpenAI benefits from another distribution channel while Google can earn infrastructure revenue even when customers choose a competing model.

Google doesn't have to win every model comparison to build an enormously successful AI business.

Google meets people at work, but Meta meets them everywhere else—arguably an even more powerful advantage.