Argon’s cost advantage over Astra depends on a 50% launch discount, whose end date Google has not confirmed. At standard pricing, its benchmark cost rises to $3.98 per task, about 20% above Astra. The savings come from lower token prices rather than lower usage: Argon averaged 62,000 output tokens per task, compared with Astra’s 27,000. Introductory prices are $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the promotion.
On the AA-Omniscience test, Argon had a 15% hallucination rate versus Astra’s 51%, reflecting a greater willingness to acknowledge uncertainty rather than guess incorrectly. Its answer accuracy was lower, however, at 50% versus Astra’s 63%. Argon also led AutomationBench-AA with 77.5%, ahead of Claude Sonnet 5.5’s 71.3%. Access remains limited to selected testers and trusted cyber defenders through Google’s Fairwind Program, alongside internal use, ahead of a broader release.