When was Gemini 4 released, and which versions can you use today?
Google has officially announced Gemini 4, but the only publicly confirmed model variant I could verify is Gemini 4 Argon. It appears to have had a phased rollout: first limited access for trusted cyber defenders, then broader availability in the Gemini app and Gemini API. Public evidence for other Gemini 4 variants like Pro, Flash, Nano, or Ultra is not yet confirmed in the primary sources reviewed.
Overview
Current evidence
Confirmed
Gemini 4 Argonthe only clearly confirmed public variant
Not confirmed
Gemini 4 Pro
Gemini 4 Flash
Gemini 4 Nano
Gemini 4 Ultra
Gemini 4 Flash-Lite
Drawn by consens.io from this answer of Oct 14, 2026. Every date, number and name is taken from it.
1) Confirmed announcements
Google’s official announcement introduced Gemini 4 Argon as the new flagship Gemini 4 model.
Google DeepMind also published a model card / model page describing Argon in more technical detail.
Google later posted that Argon was available more broadly for users and developers.
2) Release date
Timeline
Gemini 4 rollout
13 days
Public announcement of Gemini 4 Argon
Broader availability announcedbroader availability announced for Gemini app users and API users
Drawn by consens.io from this answer of Oct 14, 2026. Every date, number and name is taken from it.
The release is best described as phased, not a single all-at-once launch:
2026-09-30: public announcement of Gemini 4 Argon
Initial rollout: limited access via the Fairwind Program for trusted cyber defenders
2026-10-13: broader availability announced for Gemini app users and API users
So if you need one practical date for public availability, October 13, 2026 is the clearest milestone.
3) Model variants
Confirmed
Gemini 4 Argon (gemini-4-argon)
Not yet confirmed in official sources
Gemini 4 Pro
Gemini 4 Flash
Gemini 4 Nano
Gemini 4 Ultra
Gemini 4 Flash-Lite
Those names may be speculation or future possibilities, but I did not find official confirmation for them.
4) Technical profile
The verified docs describe Argon as:
natively multimodal: text, image, audio, and video input
text output
large sparse mixture-of-experts transformer
up to 1 million tokens of combined context / decode capacity, depending on how the docs phrase it
5) Benchmark results
The strongest publicly confirmed benchmark-type claim I found is Google’s own reporting that Argon:
ties for #1 on CWE-Bench v1 at 68% for vulnerability remediation
Google also claims strong results in applied domains like:
coding
finance
legal work
cybersecurity
science
However, I did not find a broad set of independent third-party benchmark numbers yet, so the clearest evidence is still Google’s own published evaluation and case-study results.
6) Pricing
The confirmed API pricing announced for Argon is:
$2 per 1M input tokens
$10 per 1M output tokens
cached input tokens at 95% off input pricing
That implies roughly $0.10 per 1M cached input tokens.
One source set suggests these are introductory rates that later move to $4 / $20 per 1M tokens, while another source set does not clearly confirm the end date; so the safest reading is: intro pricing is clearly confirmed, but the long-term pricing schedule should be checked live before production use.
7) Availability
Confirmed availability includes:
Gemini API for paid-tier users / GA access
Google AI Studio
Gemini app for Google AI Pro and Google AI Ultra subscribers
Google Antigravity access for certain tiers
Fairwind Program for trusted cyber-defense partners
Bottom line
The current evidence supports this summary:
Gemini 4 is real and officially announced, but the only clearly confirmed public variant is Gemini 4 Argon.
It launched in stages, became broadly available around October 13, 2026, has 1M-token-class context, and is priced at $2/$10 per million input/output tokens at intro rates.
The most concrete benchmark claim currently public is 68% on CWE-Bench v1, while broader independent benchmark data and additional Gemini 4 variants remain unconfirmed.
The standing answer said broad public release had not been announced; the new answer reports broader Gemini app and developer/API availability announced October 13, 2026. It also adds technical details and is less certain about post-introductory pricing.
The new answer reports an official Gemini 4 Argon announcement, restricted rollout, published benchmark results and API pricing—facts absent from the standing answer, which said there were no official scores or pricing and no announced variants.
Old consensus said Gemini 4 was only in pre-training with no release timeline. New consensus confirms it has advanced to post-training, with internal testing and a target for release before year-end, though still no official date, variants, benchmarks, or pricing.
OpenAI gpt-5.6-luna · Google Gemini gemini-3.5-flash-lite · DeepSeek deepseek-v4-flash
Contradiction · minor detail
Argon’s standard API prices of $4/$20 per million tokens and their effective date are confirmed.
Model B says the standard prices are confirmed to double, without specifying an effective date.
Gemini
“**Standard Pricing:** Confirmed to double to **$4.00 per million input tokens** and **$20.00 per million output tokens**.”
Model C gives the $4/$20 rates and says they begin after November 30, 2026.
DeepSeek
“Introductory Price (through Nov 30, 2026) | Standard Price (after Nov 30, 2026)”
How to verify: Check Google’s official pricing terms for whether the standard rates and November 30 transition date were confirmed.
Contradiction · minor detail
Argon has reported benchmark results beyond the CWE-Bench result, including a 77.9% DeepSWE score and a 0.7% injection-attack success rate.
Model B cites additional coding and security benchmark results.
Gemini
“reaching 77.9% on DeepSWE v1.1”
Model A says the public evidence reviewed lacks a broad independent benchmark set and treats Google’s CWE-Bench result as the clearest numerical result.
OpenAI
“The public material reviewed does not yet provide a comprehensive, independently verifiable benchmark table”
How to verify: Check the original DeepSWE and Gray Swan evaluations and whether their Argon results were public by the stated date.
Emphasis
The Vals Index is a top benchmark result for Argon, potentially more prominent than the CWE-Bench result.
Model C highlights a claimed number-one Vals Index result and top performance across several domains.
DeepSeek
“**#1 on the Vals Index**”
The consensus foregrounds the CWE-Bench score as the strongest publicly confirmed benchmark-type claim.
OpenAI
“This is the clearest numerical benchmark result publicly highlighted by Google in the sources reviewed.”
How to verify: Compare Google’s published Vals and CWE-Bench results, including their scope and whether either is independently evaluated.
