Gemini 4 Release Date: What Is Actually Confirmed?

Confirmed announcement

• Google has officially announced Gemini 4 Argon as the first model in the Gemini 4 line.
• The announcement date was September 30, 2026.
• Google frames it as a frontier model aimed at:
  • long-horizon reasoning,
  • complex software engineering,
  • enterprise knowledge work,
  • and defensive cybersecurity.

Release date / rollout status

• There is no confirmed general public release date yet.
• The model is in a phased rollout:
  • Current access: a restricted group of trusted cybersecurity defenders and internal teams.
  • Next planned access: paid API customers and Google AI Ultra subscribers.
  • General availability: not dated.

So, if you mean “when can most users use it?”, the answer is: not yet announced.

Model variants

• The only confirmed Gemini 4 variant so far is Gemini 4 Argon.
• No official evidence currently confirms additional Gemini 4 variants such as:
  • Pro,
  • Flash,
  • Nano,
  • Ultra,
  • or Lite.

One detail worth noting: Google’s public materials emphasize Argon’s cybersecurity-focused deployment mode, but that appears to be a specialized access/configuration of Argon rather than a separately confirmed public model family.

Benchmark results

Google has published several benchmark claims for Argon. The strongest confirmed ones include:

• DeepSWE v1.1: 77.9%
• Vals Index: leading performance in the published comparison
• CWE-bench v1: 68%, tied for first in the reported table
• Prompt injection robustness: Google says it is the most resilient model yet against indirect prompt injection

Important caveat: these are Google-reported results, and while they are the current official numbers, independent replication is not yet established in the evidence reviewed.

Also, the benchmark picture is not uniformly dominant across every category; in some published comparisons, Argon trails competing frontier models on certain agentic, science, or computer-use tasks.

Pricing

Google has announced introductory API pricing of:

• $2 per 1 million input tokens
• $10 per 1 million output tokens
• Cached input tokens: 95% off the input-token price

That implies cached input is roughly $0.10 per 1 million tokens at the introductory rate.

Google also says the standard post-introductory pricing will be:

• $4 per 1 million input tokens
• $20 per 1 million output tokens

Availability

Current confirmed availability is limited:

• Trusted cybersecurity partners / defenders: yes, via the restricted early-access program
• Public Gemini API / AI Studio: not confirmed as generally available yet
• Vertex AI / Google Cloud: no confirmed public listing in the evidence reviewed
• Consumer Gemini app: no confirmed release date yet

Bottom line

The best-supported current summary is:

Gemini 4 has been officially announced only as Gemini 4 Argon, which is currently a restricted-access frontier model. Google has confirmed pricing and some strong benchmark results, but has not yet announced a broad public release date or any additional Gemini 4 variants.