Reading
Gemini 4 Argon Is Half Opus's Price, but Only for Now
Argon wins 12 of Google's 18 benchmarks at half Opus 5.5's API price. Independent scores and the pricing fine print say: don't switch yet.
Prices last read 2026-10-09
Short answer
Not yet. Most people can't use Argon, and its "half the price" is an introductory rate.Google announced Gemini 4 Argon on September 30. VentureBeat reports it leads outright on 12 of the 18 benchmarks Google disclosed. Its introductory API price is $2 input and $10 output per million tokens, half of Claude Opus 5.5's $4/$20. After the introductory period, Argon's price rises to $4/$20, the same as Opus. Access started only with vetted cyber defenders in Google's Fairwind Program. Paid API customers and Google AI Ultra subscribers are next, with no date given. On October 9 we found no Argon entry in Google's Gemini API release notes or on its pricing page.Source: Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic, but in limited release (VentureBeat) · 2026-10-09; Google announces Gemini 4 Argon as its new frontier model (9to5Google) · 2026-10-09; Gemini API release notes (Google AI for Developers) · 2026-10-09; Gemini API pricing (Google AI for Developers) · 2026-10-09
What Google actually announced
According to VentureBeat's tally, Argon leads outright on 12 of Google's 18 disclosed benchmarks and ties for first on one more. GPT-6 Astra leads outright on three, and Opus 5.5 on two. Some of Argon's margins are large. It scores 51.3% on AutomationBench, against 42.5% for Opus 5.5, and 84.2% on GraphWalks, against 71.8% for Astra. On DeepSWE v1.1, which tests coding, it scores 77.9%, while Opus and Astra both score about 74%. Google also claims "industry-leading 1 million output tokens," up from 64,000. TechCrunch reports that Google pitches Argon first as a cyber-defense model that can find, validate and patch vulnerabilities.
Source: Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic, but in limited release (VentureBeat) · 2026-10-09; Google releases Gemini 4 Argon, called its most powerful model yet (TechCrunch) · 2026-10-09
The two numbers Google's chart doesn't lead with
The first is Terminal-Bench. VentureBeat's figures put Opus 5.5 ahead on Terminal-bench 4.0, 66.4% to 57.4%. Astra beats Argon on Terminal-Bench Science and FrontierSWE v2. The second comes from an independent tester, Artificial Analysis, whose Intelligence Index scores Gemini 4 Argon (High) at 53 and Claude Opus 5.5 (Max) at 58. That same page also shows a cost advantage for Argon: $1.99 per task against $5.98 for Opus at max effort.
Source: Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic, but in limited release (VentureBeat) · 2026-10-09; Gemini 4 Argon vs Claude Opus 5.5: model comparison (Artificial Analysis) · 2026-10-09
Our take: "12 of 18" is a real win on a scoreboard Google picked. Google chose those 18 tests. When a neutral tester ran its own mix, Opus came out ahead. Opus also wins Terminal-Bench, which is the closest thing here to "do real work in a shell." That matters more to most coding users than a legal-agent benchmark where every model scored under 20%. We think Argon's value is underrated, though. At the same intelligence tier, a model that costs about a third as much per task is the story worth arguing about, more than who tops the chart.
Read the price fine print before you migrate
Both VentureBeat and 9to5Google report that $2/$10 is introductory and rises to $4/$20 afterward. Neither report says how long the introductory period lasts. Cached input is discounted 95%. Google's general Gemini API terms also allow it to change pricing, as quoted below. Because Argon isn't on Google's pricing page yet, there is no official dated schedule to plan around.
Source: Google announces Gemini 4 Argon as its new frontier model (9to5Google) · 2026-10-09; Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic, but in limited release (VentureBeat) · 2026-10-09; Gemini API pricing (Google AI for Developers) · 2026-10-09
Google may make changes to this pricing from time to time
Gemini API Additional Terms of Service (last updated 2026-04-28), read 2026-10-09Source: Gemini API Additional Terms of Service · 2026-10-09
Why this matters: these are Google's general paid-API terms, not Argon-specific ones. They confirm that a listed rate can change. The same terms say changes to existing paid services usually take effect 30 days after posting, while pricing for new paid services can take effect immediately.
Source: Gemini API Additional Terms of Service · 2026-10-09
Who is affected, and what to do
Where you stand today
- Cyber defenders at governments, critical-infrastructure operators or core tech platforms: Google DeepMind's Fairwind page lists you as priority applicants. Applicants are vetted, including background checks.
- Google AI Ultra subscribers: you are in the next group, but on October 9 Google's plans page listed no Argon for any tier.
- Paid Gemini API customers: also next in line. Argon wasn't in the API release notes or on the pricing page when we checked.
- Claude and ChatGPT subscribers: nothing has changed for your plans. No Argon access exists yet to switch to.
Source: Fairwind Program (Google DeepMind) · 2026-10-09; Google AI plans (Gemini) · 2026-10-09; Gemini API release notes (Google AI for Developers) · 2026-10-09; Google announces Gemini 4 Argon as its new frontier model (9to5Google) · 2026-10-09
- Don't cancel anything todayThere is nothing to switch to yet. Cancelling Claude or ChatGPT now just leaves you without a frontier model.
- Test on your own work once you have accessPick 20 real tasks you already pay for, such as code fixes, long documents or agent runs. Compare quality and cost per task, not each vendor's headline chart.
- Plan around the full priceEstimate costs at $4/$20, not $2/$10. If Argon only wins at the introductory rate, it's a promotion, not a migration plan.
- Shell-heavy coders: Opus still has the edgeOn the Terminal-Bench figures in this report, Opus 5.5 is ahead, and that is the benchmark closest to agentic terminal work.
Our verdict
Argon is the most interesting value story of the fall, and its 1M-token output is a real capability jump. Still, a benchmark lead on Google's own test list, an introductory price and access limited to a vetted program don't add up to a reason to switch. Check back when Ultra and the API actually get it.
Optional: see seller listings for Gemini plans on SlotPrices. These are third-party listings, not official Google access, and not an endorsement.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.
Sources (all read 2026-10-09)
- Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic, but in limited release (VentureBeat)
- Google announces Gemini 4 Argon as its new frontier model (9to5Google)
- Google releases Gemini 4 Argon, called its most powerful model yet (TechCrunch)
- Gemini 4 Argon vs Claude Opus 5.5: model comparison (Artificial Analysis)
- Fairwind Program (Google DeepMind)
- Gemini API release notes (Google AI for Developers)
- Gemini API pricing (Google AI for Developers)
- Gemini API Additional Terms of Service
- Google AI plans (Gemini)