Skip to content
← Models
ClosedGoogle/10 trustVerified

Gemini 4 Argon

Google's new top-tier Gemini model for complex knowledge work, with a 1M-token output limit and the lowest measured hallucination rate in its tier.

Visit official page →

About

Gemini 4 Argon anchors Google's Gemini 4 generation after months of delays — larger than previous Pro models and, per Google, comparable to GPT-6 Astra and Claude Opus 5 on coding and cyber benchmarks. Standout features: a 1M-token max output limit and a 15% hallucination rate on AA-Omniscience, the lowest of any model scoring above 45 on the Artificial Analysis Index.

As of Oct 2026 it is announced but not generally callable — no general API availability or timeline, and the input context spec is undisclosed. Developers cannot actually call it yet.

Specs

Context window

1M tokens per third-party model listings; Google emphasizes a 1-million-token maximum output limit (input context spec not officially disclosed)

Pricing

$2 input / $10 output per 1M tokens introductory, rising to $4 / $20 after the promotion.

Free quota

None announced — no free public tier at launch (per independent coverage, Oct 3, 2026).

API access

Gemini API / Google AI Studio (paid customers first; vetted cyber partners via the Fairwind Program at launch); Google Cloud Vertex AI planned

Weights

Closed — API only

Released

September 30, 2026

Benchmarks

Google-reported:

  • DeepSWE v1.1: 77.9%
  • AutomationBench: 51.3%
  • Terminal-Bench 4.0: 57.4%
  • GraphWalks 256K–1M: 84.2%

Independent: Artificial Analysis Intelligence Index 53 (matches GPT-6 Astra); AA-Omniscience hallucination rate 15%.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

chatcodevisionvideoreasoningautomationdata

Trust score — 7/10

/10 trustVerified

Google is an established provider, and the model now has independent measurements (AA Index 53, low hallucination rate) backing the launch claims. But as of Oct 2026 it still has no general API availability or timeline, no free tier, and its input context spec is undisclosed — developers cannot actually call it yet. Scores at announcement were provider-only.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
5.1/10
Manual review · sanity read + red-flag check
7.0/10
  • · provider-tier estimate — thinner evidence (no public adoption metric)
  • · released 6 days ago
  • · manual review: announced, not callable; 16 benchmarks unpublished
  • · capped at 7: key claims are provider-only, not yet independently verifiable

Sources