Why Google and OpenAI Are Naming Models After Stars and Elements

The Naming Strategy Behind the Latest AI Models

Google and OpenAI just released new frontier models with names that seem arbitrary at first glance — until you realize they’re part of a coordinated repositioning strategy.

Google’s Gemini 4 Argon, announced September 30, sits alongside the company’s existing gas-named models. OpenAI’s GPT-6.1 Sol, revealed September 29 at DevDay 2026, follows its solar system naming hierarchy: Astra is the flagship, Sol is the mid-tier, and whatever comes next will likely occupy the outer planets. These aren’t decorative labels. They’re signaling where each company thinks the market is heading.

What Argon and Sol Actually Do

Gemini 4 Argon is engineered for sustained reasoning across long-running workflows. It handles software development, legal and financial knowledge work, and cybersecurity — the tasks that traditionally required expensive human experts. The model supports output up to 1 million tokens, a massive jump from the previous 64,000-token ceiling. Google says thousands of employees are already using it internally, including for migrating C/C++ code to Rust across the 800,000-line Zircon kernel in Fuchsia.

On benchmarks, Argon scores 77.9 percent on DeepSWE v1.1, leads AutomationBench at 51.3 percent, and ties for first on CWE-bench v1 at 68 percent. Google is also rolling it out through its Fairwind Program to trusted cybersecurity researchers before wider release.

GPT-6.1 Sol is positioned as the practical choice — powerful enough for serious work but priced for scale. Its API costs $2 per million input tokens and $10 per million output tokens, roughly one-fifth of OpenAI’s top-tier GPT-6 Astra. OpenAI claims it matches Astra’s DeepSWE v1.1 score at that lower price point and beats Anthropic’s Claude Opus 5.5 on AutomationBench by 2.2 points at about one-third the cost. A faster variant, GPT-6.1 Sol Ultrafast, is coming within days and promises up to eight times the token generation speed.

The Real Story: Pricing as Weapon

Both models share the same pricing structure — $2 in, $10 out — and both undercut their respective flagships significantly. This is the defining move of this cycle. The race is no longer just about who scores highest on benchmarks. It’s about who can deliver frontier-level performance at a price that makes enterprise adoption rational.

OpenAI’s strategy is particularly sharp. By positioning Sol as the cost-effective alternative to Astra, it’s creating a ladder that captures users who don’t need maximum capability but still want top-tier results. The Ultrafast variant is an insurance policy — it ensures that even speed-sensitive workloads stay within the Sol ecosystem rather than pushing users toward Astra or competitors.

Google is playing a similar game with Argon, but through a different channel. The 95 percent discount on cached input tokens is a structural incentive for companies running repeated or iterative workloads. If your application frequently re-processes the same context, Argon becomes dramatically cheaper over time. That’s a design choice aimed squarely at API-heavy enterprises.

Who Wins, Who Loses

Anthropic is the obvious loser in this round. Claude Opus 5.5 is already being compared unfavorably to Sol on AutomationBench at a fraction of the cost. Sonnet 5.5, also mentioned in the same coverage, doesn’t command the same premium positioning. Anthropic needs to respond — either by improving its cost curve or by finding a differentiation path that pricing alone can’t erase.

Mid-tier model consumers win. Developers and companies that previously had to choose between expensive frontier models and cheaper but weaker alternatives now have viable middle options. The Ultrafast variant is especially significant for real-time applications where latency matters as much as accuracy.

OpenAI’s Astra buyers may feel squeezed. The new Sol model delivers near-equivalent performance at one-fifth the price. Unless Astra pulls further ahead on the hardest tasks — and OpenAI itself recommends it for the most difficult scientific work on Terminal-Bench — some enterprise customers will shift down the ladder.

Why Japan Is Covering This First

ITmedia’s coverage of both announcements appearing before major English-language outlets isn’t coincidental. Japan’s developer ecosystem is deeply integrated with these models — companies like Preferred Networks, Sony, and numerous fintech and robotics firms are early adopters of Google and OpenAI APIs. Japanese tech media often sees product announcements through the lens of implementation and commercial impact, which means they’re closer to the ground when new models ship.

This pattern will likely continue. As AI model releases become more frequent and more commercially focused, Japanese outlets with strong developer audiences will remain ahead of Western wire desks on breaking coverage of these products.

What Comes Next

Both companies are clearly positioning their mid-tier models as the workhorses of the next AI cycle. The naming conventions reinforce this — Argon as the stable, reliable element; Sol as the steady, powerful center. Neither name suggests experimental or speculative capabilities. Both suggest maturity.

Expect more variants in the weeks ahead. Google’s Cyber variant already proved the formula for security-focused releases. OpenAI’s Ultrafast variant is only the first speed-optimized option. The model families are expanding, and the pricing competition between them will intensify.

The frontier model war has shifted from raw capability to accessibility. Whoever offers the best performance-per-dollar wins the enterprise contracts. Right now, both Google and OpenAI are playing that game aggressively — and Anthropic is on the back foot.