Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Google DeepMind leader Koray Kavukcuoglu announced Argon on September 30, 2026, starting phased access for trusted cyber defenders through Fairwind.
- The Fairwind page checked October 6 still restricts early access to approved trusted partners, with conditions on use, internal teams and sharing or resale.
- The announced one-million-token figure is maximum output, not a confirmed input context or each account's quota.
- Future access is planned for developers, enterprises and consumers, starting with paid API customers and AI Ultra subscribers. Public dates, interface IDs, regions and account qualifications remain unconfirmed.
- Announced introductory prices are $2 input/$10 output per million tokens, rising to $4/$20 afterward. The introductory end date and your actual bill remain unknown.
Announced does not mean open to every account
Google published the announcement on September 30, 2026, authored by Koray Kavukcuoglu, Google DeepMind senior vice president and Google chief AI architect. It describes phased Fairwind access for trusted cyber defenders and expansion after strengthening safeguards and collecting early feedback. This establishes the announcement and rollout plan.
The Fairwind page checked October 6 still limits exclusive early access to approved trusted partners. An announcement, existing API bill or AI Ultra subscription does not prove current activation for ordinary users or enterprises. No private account access, activation time or account-specific limits were inspected.
Sustained engineering work and longer output are the main promise
Google positions Argon for prolonged, complex professional tasks in software engineering, finance, legal work and cyber defense. Teams handling large repositories, repeated tools or long complete results may find that relevant. The announcement's internal cases and evaluations are Google's claims, not our tests or guarantees for your team.
The original explicitly raises the output-token limit from 64K to one million. That is output, not input context, per-minute quota or request budget. It supplies no final parameter table covering all channels and accounts. Actual access to that limit awaits the relevant interface, authorization and limits.
Fairwind prioritizes approved defense partners
The program prioritizes government and national cyber agencies, critical-infrastructure operators and widely used software or technology platforms. Hospitals, telecoms, energy and financial networks are listed examples. Alignment does not mean automatic approval: Google checks security history and ethical operating records.
An application is not immediate activation. Google says it will review and respond to qualifying partners as quickly as possible, without a fixed processing time. Academic labs performing defensive evaluation can apply. Ordinary students and ineligible organizations should consider existing public tools rather than treating Fairwind as unrestricted trial access.
Approval still limits use and team access
Approved organizations may grant Argon access only to internal cybersecurity, incident-response or penetration-testing teams and must track employee access and usage. Conditions include user-level authentication, phishing-resistant multifactor authentication and applicable access controls. Exclusive access cannot be shared, redistributed or sold externally.
Allowed dual-use work includes authorized threat simulation, reverse engineering and malware analysis for defense or academic research. Model capability or program approval does not authorize testing arbitrary targets. Define target permission, data scope, review and records first. No application, attack/defense task or real-system operation was performed here.
Paid API and AI Ultra are planned starting groups
The September 30 announcement plans expansion to developers, enterprises and consumers, beginning with paid API customers and Google AI Ultra subscribers. It gives no complete execution date, country list, subscription tiers, organization approvals or account activation mechanism. Planned priority is not current universal access.
The public Gemini API models and release notes checked October 6 did not provide a general Argon model ID or integration entry. That absence does not prove partners lack access, but cannot justify inventing a model string, endpoint or consumer selector. Watch formal docs and account notices; do not upgrade solely for unconfirmed eligibility.
Announced pricing is a plan, not every account's current rate
Google announces introductory rates of $2 input and $10 output per million tokens, with $4/$20 afterward. It gives no introductory end date or final details for each channel, tier or enterprise contract. Label these as the announcement's pricing plan.
The announcement also describes a 95% cached-input discount relative to input pricing. That does not establish a particular cache hit, savings or retention period. Actual bills, taxes, cache rules and token use remain unknown here. The million-token output limit is not a free allowance; budget actual output, reasoning, retries and final billing rules.
Prepare evaluations while keeping working integrations
Teams can inventory evaluation tasks, sensitive data, tool permissions, quality criteria and acceptable budgets. For long tasks, define stops, human review and fallback before designing around maximum output. Keep current usable models and workflows; unconfirmed access is not a production dependency ready to switch.
After formal access and docs arrive, validate requests, output structure, tools, time, errors and cost using authorized, redacted fixed samples in isolation. A demo, announcement example or model name is not personal integration or business acceptance. This guide provides no unverified shortcuts, installation commands or API samples.
Track concrete access and billing changes
Watch formal paid-API or AI Ultra availability, regions, account conditions, model ID, parameters, introductory end date and billing rules. Reassess within the same product and account conditions; Fairwind partner access does not imply personal access, and another Google feature is not a full Argon launch.
Until then, the established facts are the September 30 announcement, current Fairwind early-access rules and expansion plan. Keep unknowns explicit for budgeting and expectations. Do not fill in unannounced deadlines or treat exits, proxies or region changes as a way to gain model eligibility.
Sources
Frequently Asked Questions
Can everyone use Argon now?
No. Confirmed early access covers approved trusted Fairwind partners. Broader developer, enterprise and consumer access is announced, with account conditions, dates and general integration still unconfirmed.
Will buying AI Ultra activate Argon immediately?
The announcement lists subscribers among future starting groups, not guaranteed immediate access. Check formal notices rather than upgrading for an unconfirmed entitlement.
Is one million tokens input context?
No. The September 30 announcement refers to maximum output, raised from 64K. Input context, account quotas and request budgets require their own docs.
What are introductory prices and their end date?
The announced plan is $2 input/$10 output per million tokens, then $4/$20. The end date and channel, contract, tax and account billing details are unverified.
How long does Fairwind approval take?
No uniform processing time or automatic approval is promised. Qualifying partners undergo eligibility, ethics and security-history review.
Can a paid Gemini API account call Argon by changing model?
No general model ID was found in the checked public list or notes. Wait for formal interface and account authorization rather than guessing an ID or reusing another model's limits.