Gemini 4 Argon is Google's new frontier model for deep reasoning across complex, long-horizon workflows. It is #1 on Arena overall and coding in the October 8, 2026 data, but it is not publicly available yet: access is limited to selected trusted cyber defenders through the Fairwind Program. If you searched for gemini 4 argon release date or the reversed gemini argon 4, the short answer is: announced September 30, 2026, with no public release date yet.
The distinction matters when choosing a model. Arena measures human preference, Google's benchmarks report performance under its evaluation conditions, and an announcement does not establish public API access. This guide separates those signals using Google's official pages and dated AI Rank snapshots, checked as of October 11, 2026.
Gemini 4 Argon Arena Rankings: Overall, Coding, and Every Category
The Arena entry is gemini-4-argon-high, the high thinking variant. These text leaderboard results use style control and reflect human preference within each category, rather than a direct measure of task accuracy. The following snapshot was published October 8, 2026.
| Category | Model | Rank | Rating | 95% CI | Votes |
|---|---|---|---|---|---|
| Overall | Gemini 4 Argon high | #1 | 1525.4 | 1516.5–1534.2 | 4,892 |
| Overall | Claude Opus 5.5 high | #2 | 1507.1 | 1499.0–1515.2 | 6,272 |
| Coding | Gemini 4 Argon high | #1 | 1561.9 | 1544.7–1579.1 | 1,233 |
| Coding | Claude Fable 5 high | #2 | 1551.4 | 1544.7–1558.1 | 10,404 |
→ Track it on AI Rank: Gemini 4 Argon — Arena rating, rank and 90-day history
The overall lead is clear on this snapshot: Argon's lower confidence bound, 1516.5, exceeds Opus 5.5 high's upper bound, 1515.2. Coding deserves more caution. Argon's 1,233 votes produce a wider interval, 1544.7–1579.1, which overlaps Fable 5 high's 1544.7–1558.1. The coding lead is narrower than the point estimate suggests; the intervals do not establish a clear-cut separation.
Argon also ranks first in every category shown in the Arena detail snapshot. These categories give more context than the overall score alone, although their vote counts differ substantially.
| Category | Rank | Rating | Votes |
|---|---|---|---|
| Chinese | #1 | 1587.7 | 368 |
| Coding | #1 | 1561.9 | 1,233 |
| Creative writing | #1 | 1519.2 | 1,053 |
| Instruction following | #1 | 1528.7 | 1,750 |
| Math | #1 | 1528.3 | 271 |
| Multi-turn | #1 | 1553.3 | 670 |
| Overall | #1 | 1525.4 | 4,892 |
Math has only 271 votes and Chinese 368, so those first-place results should be read with particular caution.
→ Coding leaderboard: Arena coding rankings on AI Rank (Arena data published Oct 8, 2026).
The overall history records #1 on September 30, October 2, and October 8, with ratings of 1524.8, 1525.2, and 1525.4 respectively. That supports a lead across those recorded snapshots, without guaranteeing future rankings.
There is no OpenRouter usage ranking for Argon yet. It has no entry in OpenRouter's public model catalog and is absent from the checked day, week, and month usage rankings through October 10. Available Google models there include Gemini 3.8 Flash. OpenRouter usage on AI Rank measures activity on that platform; Argon's absence does not establish zero usage elsewhere.
What Google Reports on Benchmarks
Google reports the following results in its DeepMind Gemini benchmark table. All values below are percentages. They are vendor-reported results, and AI Rank did not independently reproduce them.
| Benchmark | Gemini 4 Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|
| Vals Index | 68.9% | 63.1% | 65.8% | 67.0% |
| AutomationBench | 51.3% | 41.4% | 31.4% | 42.5% |
| DeepSWE v1.1 | 77.9% | 74.1% | 67.4% | 74.2% |
| FrontierSWE v2 | 55.0% | 65.5% | 56.3% | 62.3% |
| Terminal-bench 4.0 | 57.4% | 58.2% | 57.9% | 66.4% |
| GraphWalks 256k–1M | 84.2% | 71.8% | 65.0% | 66.8% |
| LVBench | 91.7% | 87.5% | 79.7% | 83.7% |
| CWE-bench v1 | 68.0% | 68.0% | 58.0% | 67.0% |
| OSWorld-2.0 offline partial | 69.2% | 72.6% | — | — |
Google reports leads for Argon on several rows, including AutomationBench and DeepSWE v1.1, but the picture is mixed. Argon trails all three listed competitors on FrontierSWE v2 and Terminal-bench 4.0, trails GPT-6 Astra on OSWorld-2.0 offline partial, and ties Astra on CWE-bench v1. Dashes mean Google's table lists no value.
Google's evaluation methodology says Argon uses the highest thinking settings, with scores reported as pass@1 unless noted. Competitor results are mostly providers' self-reported figures; some results are computed by Google, including Argon's DeepSWE and Terminal-bench 4.0 scores. These conditions limit how confidently the table predicts performance in a particular deployment.
Gemini 4 Argon Release Date and Availability
Google announced Argon on September 30, 2026, describing uses in software engineering, enterprise knowledge work such as legal and finance, and cybersecurity defense. The initial rollout is through Fairwind, a program with more than 650 partners globally. Only a selected set receives Argon access, standalone or within CodeMender; that partner total is not a count of Argon users.
Fairwind prioritizes governments and national cyber authorities, critical infrastructure operators, and core technology platforms. Organizations can apply through its form. Access requires user-level authentication, phishing-resistant MFA, and tracked use, and is restricted to internal cybersecurity, incident response, and penetration testing teams. Sharing, redistribution, and resale are prohibited. Zero data retention is supported when accessed as a managed model on Gemini Enterprise.
Google says broader access will start with paid API customers and Google AI Ultra subscribers, while it participates in the U.S. government's voluntary pre-release model access process. No date is given. As of October 11, there is no public Gemini API model ID or Argon entry on the Gemini API pricing page.
Google also announces a 1M-token output limit, up from 64K. That is an output limit; the announcement does not state an input context window.
Announced Pricing (Not Yet Payable)
The Google announcement sets out the following planned API prices. These are announced prices, not yet payable public API rates as of October 11, 2026.
| Announced period | Input per 1M tokens | Output per 1M tokens |
|---|---|---|
| Introductory | $2 | $10 |
| After the introductory period | $4 | $20 |
Google announces cached input at 95% off the input price. The end of the introductory period is not dated, and Google has not announced batch or long-context pricing tiers. Use these figures for provisional planning, and check the public pricing entry when access becomes available.
Limitations and What to Check
The strongest supported conclusion is that Argon leads this dated Arena snapshot, while Google's benchmark table shows strengths and losses across different tasks. Neither source establishes how it will perform in your workflow. The high thinking Arena variant, unequal vote counts, and vendor evaluation settings all matter when comparing models.
Before planning a deployment, check for a public release date, a documented Gemini API model ID, and a live pricing entry. Confirm the introductory period's terms and distinguish output capacity from input context. For ranking comparisons, use the latest votes and intervals; for availability, rely on Google's access documentation.
Compare before you choose: rankings move as new votes come in — check the current Arena top-rated models on AI Rank before you commit.
Or 📱 get the AI Rank app to follow them on your phone.
Prepared with automated editorial assistance and checked against official sources and dated AI Rank data on October 11, 2026. We did not run our own benchmarks for this article.