Gemini 4 Argon: Release Date, Pricing, API and How to Use It

Gemini 4 Argon is Google DeepMind’s first Gemini 4 model: a frontier reasoning model for coding, agents and knowledge work, announced on September 30, 2026 and opened to everyone on October 13, 2026. When this guide was first published, Argon was limited to Fairwind Program testers and had no public model ID. That has changed. The model is now listed in the Gemini API as gemini-4-argon, has a row on the official pricing page, and is available in the Gemini app, AI Mode in Search and Google Antigravity for paying subscribers.

This October 30 refresh adds the general-availability details from Google’s October 13 announcement, the Gemini API developer guide and pricing page (last updated October 27 and 28, 2026), Google Cloud Agent Platform pricing, and independent results from Artificial Analysis. The September 30 launch material stays below for the vendor-reported benchmark tables and Google’s internal case studies, with outdated access statements corrected.

What Google announced

The lead announcement comes from the Google DeepMind account and frames Argon as a frontier model rather than a refresh of an existing tier. The stated target workloads — complex workflows across coding, enterprise knowledge work, and cybersecurity defense — describe long-running agentic work rather than chat. The distribution method is the part that differs most from a normal Gemini release: a set of trusted testers, reached through the Fairwind Program, on day one.

A follow-up from the same account addresses the output ceiling directly, and a longer thread from Sundar Pichai gives the framing Google wants attached to the release: broad internal use at Google already, with feedback from teams spanning coding through quantum computing.

Google’s corporate account repeats the three workload areas and states the output limit as a differentiator. Read the two texts side by side and the emphasis is consistent: the same three domains, the same 1M number, the same “frontier” label.

That last post is worth reading carefully because it sets the evidence standard for everything that follows. Google is describing its own internal usage and its own benchmark runs. Pichai’s post says “here’s a look at the benchmarks” and attaches charts, which makes those charts primary evidence for the claims in this article — but they remain vendor-run measurements, not independent ones.

Context window and the 1M token output limit

Output limits are the constraint that quietly shapes agent architecture. When a model can only emit 64K tokens in one response, long-horizon work has to be chopped into steps, state has to be handed between calls, and the orchestration layer carries the burden of stitching a result back together. Raising the ceiling to 1 million tokens changes where that burden sits — a single generation can carry work that previously required a loop.

The official numbers for comparison, stated on Google’s own channels:

How the 1M output works in practice: the model generates in hops of up to 64K tokens, and each hop continues from where the last one stopped. With background: true and a max_output_tokens value above 65,536, Google’s servers run those continuation hops for you, and you join the text of every model_output step. If you turn on background execution without setting max_output_tokens, the server stops at 64K. Google recommends 131,072 or 262,144 for complex reasoning and coding, and at least 32,768 for apps that used a lower cap, because Argon spends more of the budget on thinking than earlier models.

Two cost details matter for planning. Output pricing includes thinking tokens, so a high thinking level raises the bill even when the visible answer is short. And Google Cloud’s Agent Platform pricing says that from November 13, 2026, completions longer than 32K tokens are billed for retrievable token usage even if the request later fails on a client or server timeout. Google’s developer guide recommends HTTP client timeouts of at least 15 minutes, or background execution, for hard prompts.

Gemini 4 Argon benchmarks (vendor-reported)

Google’s announcement leads with benchmark gains across coding, general agentic capability, and long-context video understanding. Every number in this section is vendor-reported: Google selected the comparison models, ran or commissioned the evaluations, and published the charts. No independent reproduction was found in first-party sources on announcement day, and no harness configuration or run log was published alongside these figures.

One row deserves explicit framing. CWE-bench v1 is not a clean win: the published chart shows a three-way tie at 68% involving Argon, Grok 4.7, and GPT-6 Astra. Google lists Argon as tied for first, which is accurate and also means the security-patching result does not separate Argon from the field. If a decision depends on that specific benchmark, the honest conclusion is a draw, not a lead.

The same idea applies to the benchmark set as a whole. The eval mix Google chose emphasizes software engineering, automation, video understanding, and security. That is a reasonable proxy for the three workload areas in the announcement, but it is not a general capability index, and a model can look strong on this particular selection while being unremarkable on reasoning or multilingual work.

Independent benchmarks after release

Since general availability, outside testers can run Argon directly. Artificial Analysis scored Gemini 4 Argon (high reasoning) at 53 on its earlier Intelligence Index at launch, level with GPT-6 Astra (max) and one point ahead of GPT-6.1 Sol (max). On the current Intelligence Index v5.0 it scores 50, ranked 13 of 57 models when checked on October 30, 2026.

The independent numbers support Google’s agentic and hallucination claims more than a general intelligence lead. Argon sits in the leading cluster but not at the top, and at high reasoning it is slow to start answering and heavy on tokens. For interactive use, test medium or low thinking before defaulting to high.

Cyber defense evaluations

The cybersecurity results are where the announcement spends the most chart space, which matches the fact that the first access cohort is cyber defenders. Two of the charts cover public benchmarks; two cover Google-internal or partner-run security work.

CWE-bench measures whether a model can patch real vulnerability classes. A 68% result is genuine progress on a hard benchmark, and the chart makes the tie visible rather than burying it — which is the useful part for a reader making a decision.

The Gray Swan chart reads in the opposite direction from the others: the metric is attack success rate, so a lower number is a better result. Google reports Argon at a 0.7% attack success rate at k=15 attempts, the strongest profile in the comparison set. For anyone deploying agents that read untrusted content — web pages, issue trackers, email — indirect prompt injection robustness is one of the few security properties with a directly measurable number, so this chart is more decision-relevant than the coding scores.

The comparison target here is instructive. Google benchmarks Argon against 3.8 Flash Cyber, its own cyber-tuned model, rather than only against competitors. Beating an internal specialist model by roughly 15 points on vulnerability discovery is a meaningful internal progression. It is also an internal comparison, evaluated on internal infrastructure, with no published test set, and the reader should treat the absolute percentages as directional. The Wiz row is different in kind: Wiz is an external security vendor and the benchmark is theirs, but Google is still the party reporting the result, and no Wiz write-up of the same run was located in first-party sources.

What Google ran internally

The most distinctive part of the announcement is a set of internal case studies, because Google is unusually specific about the work rather than just claiming productivity gains. These are internal results with no external test harness, so they illustrate capability rather than prove it — but they are concrete enough to reason about.

The migration numbers connect most directly to the output limit. Replacing 32K lines of SIMD code, or moving 800K+ lines of kernel code, is the class of task that a 64K output ceiling makes awkward and a 1M ceiling makes feasible in fewer passes. That relationship is a plausible mechanism rather than proof, because Google has not published the prompt structure, number of turns, or review process behind either migration.

Treat the qualitative wording carefully. “2.7× faster than the Rust port” is a relative claim between two internal artifacts, and “closer to optimized C++” is Google’s own characterization rather than a benchmark result. The quantum-optimization figure has no published baseline reference or evaluation protocol, so it is best read as a signal of where Google is applying the model rather than as a performance claim you can compare against anything.

Release date and rollout

At launch, access ran only through the Fairwind Program rather than the Gemini API; general availability followed on October 13, 2026 (see the table below). Fairwind was introduced on September 2, 2026 as a defender-focused program; the DeepMind program page describes it as working with a global network of partners, and reports 650+ partners worldwide. That figure covers the program as a whole, not Argon access specifically.

The Fairwind program page also connects earlier DeepMind security work to this release, naming Gemini 3.8 Flash Cyber and CodeMender, the code-security agent DeepMind published in October 2025. Argon is the third step in that line: a cyber-tuned model, an automated patching agent, and now a frontier model whose first cohort is defenders.

General availability came 13 days after the first reveal, as VentureBeat reported. At launch, Google gave no dates for the later stages. Both have now happened, and consumer access went to Google AI Pro as well as Ultra subscribers.

Gemini 4 Argon pricing: API and plans

Google now publishes Argon pricing on the official Gemini API pricing page (last updated October 28, 2026). The introductory period has a fixed end date: November 30, 2026. Gemini 4 Argon is not available on the Gemini API free tier.

On Google Cloud’s Gemini Enterprise Agent Platform, global endpoints list the same $2 / $10 introductory and $4 / $20 standard rates, with non-global endpoints about 10% higher. There, the introductory discount is paid as 50% credits back on net spend. A separate Gemini 4 Argon Cyber Permissive model for approved defenders costs $4 / $20 with no promotion.

For budgets, build cost models on $4 / $20, because that is the price from December 1. Explicit context caching is not supported on Argon. Caching is implicit for prompts of at least 4,096 tokens and cuts cached input by 95%, so agents that resend a stable prefix still save automatically. Long generations dominate the bill: a full 1M-token output costs about $10 at the introductory rate and $20 afterward, before input.

Consumer plans are billed separately from the API. Argon in the Gemini app and AI Mode in Search needs Google AI Pro or Ultra, and individual access in Google Antigravity needs Ultra. For how Antigravity plans, quotas and AI credits work, see our Antigravity credits and pricing guide.

How to use Gemini 4 Argon

There are four ways in, depending on whether you want a chat app, a coding agent or an API.

1. Gemini app and AI Mode in Search

1. Subscribe to Google AI Pro or Google AI Ultra.
2. Open the Gemini app or AI Mode in Google Search while signed in to that account.
3. Choose Argon in the model menu if it is not already selected. Google says Argon is available there for Pro and Ultra subscribers from October 13, 2026; menu labels can change by region and device.

2. Google Antigravity

Google AI Ultra subscribers and enterprise customers can select Argon as the model in Google Antigravity, Google’s agent-first development platform. Google says Argon in Antigravity raised agentic code submissions inside Google by more than 35% while cutting rollbacks by nearly a third (vendor-reported). Through the API, the Antigravity agent in Gemini Managed Agents also accepts gemini-4-argon as its model.

3. Google AI Studio

Open Google AI Studio, pick Gemini 4 Argon, and prompt it in the playground. Google lists AI Studio access for developers on a paid-tier API project and for Google AI Ultra subscribers.

4. Gemini API

1. Create an API key in AI Studio on a project with billing enabled. Argon has no free tier.
2. Install the Google Gen AI SDK: pip install google-genai or npm install @google/genai.
3. Call gemini-4-argon:

from google import genai

client = genai.Client()  # reads GEMINI_API_KEY (paid tier)

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Review this repository layout and propose a migration plan.",
)

print(interaction.output_text)

For outputs longer than 64K tokens, use background execution and join every output step:

import time
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-argon",
    input="Write the full test suite for the attached module.",
    background=True,
    generation_config={
        "max_output_tokens": 262144,   # above 65,536 enables long decode
        "thinking_level": "high",      # low | medium (default) | high
    },
)

while interaction.status in ("in_progress", "queued"):
    time.sleep(5)
    interaction = client.interactions.get(id=interaction.id)

# each continuation hop adds a model_output step: join them all
text = "".join(
    c.text
    for step in interaction.steps if step.type == "model_output"
    for c in step.content if c.type == "text"
)

Migration notes from Google’s developer guide: remove temperature, top_p and top_k (Argon ignores them); drop explicit cachedContents calls; keep multi-turn state with previous_interaction_id; and test all three thinking levels, then choose the lowest that meets your quality bar. For cheaper high-volume work, Gemini 3.8 Flash is still the lower-cost Gemini option.

Safeguards

The rollout framing is explicitly cautious. Access begins with a narrow cohort of defenders, which is itself a risk-limiting decision: security work has clear legitimate uses, measurable outcomes, and partners who can report misuse. Google’s Frontier Safety Framework pages describe the company’s staged evaluation approach for frontier models, and the announcement points to that framework as the governance context.

Google has since published a Gemini 4 Argon model card, linked from its October 13 announcement, which closes the gap this section flagged at launch. Google reports that external bio red teaming unlocked fewer than 1% of the bio sub-tasks it tests, that shortcut attempts on coding tasks fell to 2%, and that attack success on Gray Swan indirect prompt injection was 0.7% across 15 adaptive attempts. All three figures are vendor-reported. Cyber defenders who need fewer refusals can apply through Fairwind for a version without cyber guardrails.

What this changes for agent workflows

Setting the access question aside, the capabilities Google emphasizes point at four workflow patterns that get easier if the claims hold.

Long single-pass generation. A 1M output budget makes it reasonable to ask for a complete artifact in one call — a full migration patch set, a large refactor across files, a generated test suite — instead of driving a loop of small edits. The tradeoff is cost concentration and review difficulty: a single enormous response is harder to review incrementally than ten smaller ones, so the workflow needs a verification step rather than more prompting.

Security review with measurable robustness. The indirect prompt injection result is the most actionable number in the release for agent builders. If an agent ingests web content, tickets, or email, injection robustness is a live threat model, and a 0.7% attack success rate is a figure you can reason about — while remembering it is a vendor-reported run at a fixed attempt budget.

Migration and large-codebase work. The C/C++ to Rust and Fuchsia figures describe long-horizon code transformation, which is the workload class where output limits historically forced the most orchestration overhead. Teams doing this work today typically split it across many calls; the Argon framing suggests consolidating passes.

Vulnerability triage. Combining a higher vulnerability-discovery rate with a patching benchmark (CWE-bench) describes a triage-to-fix loop. The caveat is unchanged: both numbers are vendor-reported, and the patching result is a three-way tie.

A practical adoption sequence now that Argon is public: run your own evals at low, medium and high thinking; raise max_output_tokens to at least 32,768; budget on the December prices of $4 / $20; move long jobs to background execution; and measure time to first token, which Artificial Analysis put at about 48 seconds at high reasoning.

Caveats

The launch benchmarks are vendor-reported. Google ran or commissioned them, picked the comparison models, and published the charts. Independent results arrived only after general availability (see the Artificial Analysis section), and several results — the vulnerability-discovery and quantum-optimization figures — describe internal work with no published methodology.

The security benchmark lead is not always a lead. CWE-bench v1 is a three-way tie at 68%. Google’s “tied for first” language is accurate; treating that row as a win would not be.

The API surface was missing at launch and is published now. The model ID, pricing row, developer guide and model card all arrived with general availability on October 13, 2026. Statements in the vendor benchmark section describe the September 30 launch materials.

Two of the four launch posts in the embeds above are quoted in part. The DeepMind output-limit post and Google’s derived posts continue beyond the opening clause shown; the full text is available at the linked post. Nothing here paraphrases text that could not be read directly.

Introductory pricing ends November 30, 2026. Re-run any cost estimate made in November at $4 / $20 per million tokens.

Internal results are illustrative. The memory, migration, and quantum figures show where Google is applying the model. They are not comparable to external benchmarks and, in the case of the 2.7× Rust comparison, the baseline is another internal artifact.

Use it if

Use Argon if you need Google’s strongest model for long-horizon coding, agent work, document-heavy analysis, or single outputs longer than 64K tokens, and you can accept a slow first token. November is the cheapest month to evaluate it.

Pick something else if you need low latency or the lowest cost per call. Gemini 3.8 Flash covers most high-volume work for less, and GPT-6.1 Sol lists $2 / $10 as its standard price in VentureBeat’s comparison.

Apply for special access if you are a cyber defender (Fairwind, without cyber guardrails) or a scientist (up to 10,000 free six-month seats on Antigravity’s top usage tier, subject to selection).

Earlier Guidance (September 30, 2026)

These sections come from the launch-day version of this guide. They are kept in full, with a dated update where Google’s October 13, 2026 general-availability release changed the facts.

The 1M token output limit

Output limits quietly shape agent architecture. When a model can only emit 64K tokens in one response, long-horizon work has to be chopped into steps, state has to be handed between calls, and the orchestration layer carries the burden of stitching a result back together. Raising the ceiling to 1 million tokens changes where that burden sits: a single generation can carry work that previously required a loop. Google described it as an “industry-leading 1M token output limit” that “adds a deeper level of reasoning to tackle” complex tasks.

At launch, three qualifications applied. No first-party source defined the exact semantics of the 1M ceiling, such as streaming, multi-turn continuation or per-attempt accounting. Google had not published a rate limit, latency or time-to-first-token figure. And with no model ID in the public API documentation, an outside developer could not test the claim.

Benchmark snapshot

Google’s announcement led with gains across coding, general agentic capability and long-context video understanding. Every launch figure was vendor-reported, with Google selecting the comparison models and publishing the charts: DeepSWE v1.1 at 77.9% (claimed state of the art), AutomationBench at 51.3% (ranked first), LVBench at 91.7% (state of the art), a leading Vals Index position, CWE-bench v1 at 68% (tied for first), and a leading Gray Swan indirect prompt injection profile.

CWE-bench v1 was not a clean win: Argon, Grok 4.7 and GPT-6 Astra tied at 68%. The eval mix also emphasized software engineering, automation, video understanding and security, which is a reasonable proxy for the announced workloads but not a general capability index.

Fairwind Program and rollout

At launch, access ran through the Fairwind Program rather than the Gemini API. Fairwind was introduced on September 2, 2026 as a defender-focused program, and the DeepMind page reports 650+ partners worldwide for the program as a whole. Google announced three stages: trusted testers and an initial cohort of cyber defenders on September 30; paid API customers and Google AI Ultra subscribers next; then developers, enterprises and consumers broadly. No dates were attached to stages two and three, so there was no published timeline to plan against.

Pricing

Google published introductory pricing with the announcement: $2.00 per 1M input tokens and $10.00 per 1M output tokens, with a 95% discount on the cached portion of input, stepping up to $4.00 and $20.00 afterward. There was no separate premium for cyber-defense capability. The rates were exactly double after the promotion, so cost models were best built on $4 / $20. At launch they existed only in the announcement text and charts, not on the public Gemini API pricing page.

API status today

On September 30, 2026 the announcement and the developer documentation disagreed. The Gemini API models list (last updated September 4) had no Argon model ID, the pricing page (September 8) had no Argon row, the DeepMind model card index had no Argon card, no Argon-specific safety evaluation was located, and no official demo video was found. The only access channel in the announcement was Fairwind trusted testers.

FAQ

Sources

Sources checked October 30, 2026 (launch sources first checked September 30, 2026):

• Gemini 4 Argon is now available for everyone, Google blog, October 13, 2026
• What’s new in Gemini 4 Argon, Gemini API docs
• Gemini 4 Argon developer guide, Google AI Studio
• Gemini API pricing
• Gemini Enterprise Agent Platform pricing, Google Cloud
• Gemini 4 Argon model page, Artificial Analysis
• Gemini 4 Argon launch analysis, Artificial Analysis
• Google makes Gemini 4 Argon broadly available, VentureBeat
• Introducing Gemini 4 Argon, Google blog, September 30, 2026
• Fairwind Program, Google DeepMind
• Google’s Fairwind Program, Google blog, September 2, 2026
• Gemini API models documentation
• Frontier Safety Framework, Google DeepMind
• Introducing CodeMender, Google DeepMind
• @GoogleDeepMind launch post, output-limit follow-up, @Google announcement and @sundarpichai post

Related reading: Gemini 3.8 Flash complete guide, Gemini 3.8 Live and Extended Thinking guide, Codex Security Cloud and Daybreak Blue guide, and GPT-6.1 Sol benchmarks and pricing guide.