Google's Gemini 4 Argon ties GPT-6 Astra at 53 points

Google's previous frontier model, Gemini 3.1 Pro, arrived more than seven months before Argon, and the gap was not smooth: Google skipped the already-announced Gemini 3.5 frontier model entirely during what the article calls a difficult and drawn-out development period. Argon is the result of that delay, and Google is positioning it as a return to the top tier rather than a clear win. The launch is unusually cautious — the company is not handing the model to the public but to a group of "trusted cyber defenders" in the Fairwind program and its own internal teams, with feedback from those testers feeding into safety mechanisms before developers, businesses, and consumers get access. Google has not given a date for that broader rollout, saying only "as soon as possible."

The independent benchmark picture is mixed in a way that will matter to buyers. Artificial Analysis places Argon at 53 points at its "High" reasoning level, level with GPT-6 Astra and Claude Fable 5.1 and one point ahead of GPT-6.1 Sol. Anthropic's Claude Opus 5.5 and Claude Sonnet 5.5 remain ahead at 58 and 56 points, and Argon's gains on agentic tasks — first place on AutomationBench-AA at 77.5 percent, 57 percent on Terminal Bench 4 — still leave it behind Claude Sonnet 5.5 and Claude Opus 5.5 on the latter. Google's own benchmarks show Argon leading in most cases, and Argon tops the Vals Index at 68.9 percent, but the article notes that the mid-tier Claude Sonnet 5.5 outranks Anthropic's own top model there, advising readers to take that result with a grain of salt.

What the launch ultimately hinges on is whether the price advantage holds up in practice. At the current promotional rate, one Intelligence Index task costs $1.99, which is 60 percent of GPT-6 Astra's cost but 2.7 times more expensive than GPT-6.1 Sol. Once the discount ends, the cost rises to $3.98, about 20 percent above GPT-6 Astra. The reason is that Argon burns far more tokens per task than Astra, so a low per-token price does not automatically translate into a low cost per task. For API customers and Ultra subscribers deciding what to run in production, that distinction — and how the Gemini app compares with the software wrappers around rival models — is likely to matter more than the benchmark tie itself.