Google's Gemini 4 Argon Matches GPT-6 Astra at 60% of the Cost — on a Launch Discount, and Only Cyber Defenders Can Use It Yet

GoogleModels2026-10-02not yet featured

Google's Gemini 4 Argon Matches GPT-6 Astra at 60% of the Cost — on a Launch Discount, and Only Cyber Defenders Can Use It Yet

Google unveiled Gemini 4 Argon on Sept 30, 2026, its first Gemini 4 model, with output capacity raised to 1 million tokens from 64,000 and introductory pricing of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 once the launch window ends. Artificial Analysis, the independent benchmarking firm, scores Argon at 53 on its Intelligence Index — tied with OpenAI's GPT-6 Astra — at $1.99 per Index task against roughly $3.30 for Astra, 60% of the cost. That edge is a function of the 50% launch discount rather than efficiency: Argon averages about 62,000 output tokens per task against Astra's 27,000, so at full price Artificial Analysis puts its cost per task at $3.98, about 1.2x Astra. The model is also not generally available: Google is releasing it first to trusted members of its Fairwind Program, a group of cybersecurity defenders using it to find and patch vulnerabilities, before any wider rollout.

X · Facebook · LinkedIn1200×675 Instagram · Facebook Reels1080×1920 Douyin · 抖音1080×1350

Receipts

Flagged:Primary is Artificial Analysis's own write-up of its Gemini 4 Argon evaluation — a named research firm, and the source of all three of the card's headline stats (Intelligence Index 53 tied with GPT-6 Astra; $1.99 vs ~$3.30 per task, i.e. 60%; ~62K vs ~27K output tokens per task, giving $3.98 or ~1.2x Astra once the 50% launch discount ends). Google's own launch post is the primary disclosure for the 1M-token output ceiling (up from 64K), the $2/$10 intro pricing rising to $4/$20, and the Fairwind-Program-first gating; CNBC and VentureBeat independently reported the launch and the limited-release framing. Freshness: announced Wednesday Sept 30, built Friday Oct 2 — at the 48-hour edge of rule 2, flagged to the operator at screening and approved with that caveat. Deliberately excluded (rule 9): Alphabet's Oct 1 stock move (up ~2% premarket, flat to down by midday), because Alphabet's Suncatcher TPU launch aboard a SpaceX Falcon 9 was scheduled the same day and the move cannot be cleanly attributed. Some outlets phrase the cost edge as '40% cheaper'; the card uses Artificial Analysis's own '60% of the cost per task' framing, which is the same figure. Several source domains are egress-blocked from this environment for direct fetch, so figures were cross-checked across Artificial Analysis's article and its own X post, CNBC, VentureBeat and 9to5Google rather than fetched individually; all agree.