What is the difference between Gemini 4 Argon and Gemini 3.8 Flash?

Gemini 4 Argon (announced September 30, 2026) is Google's frontier model — a new generation built for long-horizon agentic coding, legal/finance work and cyber defence, with a 1M-token output limit. Gemini 3.8 Flash (September 2, 2026) is the fast, cheap workhorse tier: $0.75/$3.75 per million tokens through December 31, 2026, 64K max output, text+image+audio+video input. Argon costs 2.7x more at its intro price and is not yet publicly available; 3.8 Flash is generally available and powers Google AI Mode.

How much do Gemini 4 Argon and Gemini 3.8 Flash cost?

Gemini 4 Argon: $2 input / $10 output per million tokens at its introductory rate, rising to $4/$20 (no date given); cached input 95% off. Gemini 3.8 Flash: $0.75/$3.75 at its intro rate through December 31, 2026, doubling to $1.50/$7.50 from January 1, 2027; Batch and Flex 50% off. On a 30K-input / 5K-output task Argon costs $0.11 at intro and $0.22 after; 3.8 Flash costs $0.041 now and $0.0825 from 2027.

Is Gemini 3.8 Flash good enough for coding?

For most day-to-day coding, yes. Gemini 3.8 Flash scores 58.7 on the Artificial Analysis Intelligence Index (September 2026) and its Cyber variant produced 2.6x more correct Chrome security patches than larger commercial models in Google's testing. But it is token-hungry — about 48,000 output tokens per AA task, versus 2,200–14,000 for GPT-6 Astra — so cheap per-token pricing does not always mean cheap per task. Argon sets a new DeepSWE v1.1 record at 77.9% and is the better choice for multi-hour migrations.

Can I use Gemini 4 Argon instead of Gemini 3.8 Flash today?

No. As of October 1, 2026, Argon is available only to vetted cyber defenders in Google's Fairwind Program; paid API customers and Google AI Ultra subscribers are next, with no date. Gemini 3.8 Flash is generally available on the Gemini API and Vertex AI. Gemini 3.8 Flash Cyber is also gated via Fairwind. Build on 3.8 Flash now and plan an Argon tier for the hardest tasks when it opens.

Quick Answer

Gemini 4 Argon vs Gemini 3.8 Flash 2026: Which Google Model

Published: October 1, 2026

The short answer

Gemini 4 Argon is Google’s new frontier model (September 30, 2026): $2/$10 intro pricing, 1M-token output, #1 on the Vals Index, gated to cyber defenders for now. Gemini 3.8 Flash is Google’s generally available workhorse (September 2, 2026): $0.75/$3.75 through December 31, 2026, 64K output, fully multimodal input. Use 3.8 Flash for everything you ship today and for high-volume, latency-sensitive work; reserve Argon for long-horizon agentic coding, legal/finance research and vulnerability remediation once it is available. Facts verified October 1, 2026.

Side by side

	Gemini 4 Argon	Gemini 3.8 Flash
Announced	September 30, 2026	September 2, 2026
Tier	Frontier (new Gemini 4 generation)	Flash (speed/cost tier)
Input / output per MTok	$2 / $10 intro → $4 / $20	$0.75 / $3.75 through Dec 31, 2026 → $1.50 / $7.50
Cached input	$0.10 (95% off)	Standard Gemini cache discount
Batch / Flex	Not stated at launch	50% off
Context	1M	1M
Max output	1M tokens (Google; Vals ran 262K)	64K
Input modalities	Text, image, long video	Text, image, audio, video
Knowledge cutoff	Not stated	March 2026
Availability	Fairwind Program only; paid API + AI Ultra next	GA on Gemini API, Vertex AI; default in AI Mode
Cyber variant	Shipped to Fairwind defenders without cyber guardrails	Gemini 3.8 Flash Cyber (Fairwind, separate model)
Vals Index	68.90%, #1 of 41	Finance Agent v2: 61.44% (#2 behind Argon)
AA Intelligence Index	Not yet scored	58.7
DeepSWE v1.1	77.9%	—
CWE-bench	v1: 68% (tied #1)	v0: frontier-level (Cyber variant)
Output tokens per AA task	—	~48,000 (high)

Prices from Google’s launch post and the Gemini API pricing page as of October 1, 2026.

What Argon adds over Flash

Output length. 1M tokens in a single trajectory versus 64K. Google says this is what lets Argon “think deeply and generate hundreds of thousands of tokens” to solve a hard problem in one go — the libgav1 SIMD-to-Rust rewrite and the 800K-line Zircon kernel migration are the examples. Flash has to be chunked and orchestrated for that scale of job.
Frontier scores. DeepSWE v1.1 77.9%, AutomationBench 51.3%, Harvey Legal Agent 19.6%, Vals Finance Agent v2 65.4% (3.8 Flash: 61.44%, the only place Google’s own table puts Flash near Argon).
Prompt-injection resistance. 0.7% attack success rate on Gray Swan’s IPI benchmark, Google’s lowest ever; the chart does not list 3.8 Flash.
Vulnerability discovery. On Wiz’s black-box pentest benchmark Argon beats 3.8 Flash Cyber at mapping attack surface, finding bugs and producing proof-of-concept evidence; it found a hospital-software vulnerability “previous frontier models had missed.”

What Flash keeps

Price. 2.7x cheaper than Argon’s intro rate and 5.3x cheaper than its post-intro rate on both input and output, plus 50% Batch/Flex. The $0.75/$3.75 rate runs through December 31, 2026 — fixed and dated, unlike Argon’s open-ended “introductory.”
Availability. GA on the Gemini API and Vertex AI since September 2, 2026, with audio input Argon’s launch post does not mention. Argon has no public date.
Latency. Vals measured Argon at 46 minutes 33 seconds average latency per Index test at high effort; that is a batch model, not a chat model. Flash is the interactive tier.
Patching track record. 3.8 Flash Cyber’s 2.6x Chrome-patch result via CodeMender stands; Argon “builds on” it, and CodeMender is listed alongside Argon in Fairwind.

Cost per task reality

Reference 30K-in / 5K-out task: Argon $0.11 (intro) / $0.22 (post); 3.8 Flash $0.041 (2026) / $0.0825 (2027). But Artificial Analysis found 3.8 Flash emits ~48K output tokens per task at high effort — 7.7x GPT-6 Astra — which erased most of its price advantage against Astra in measured cost ($2.04 vs $1.41–$3.27 per coding task). Argon’s per-task token use is unmeasured by AA; Vals’ $15.68 per Index test and $57.82 per Code Migration test suggest it uses its headroom. Run both on your workload at the effort level you will ship.

Which to choose in 2026

Shipping now, any volume: Gemini 3.8 Flash. It is the only one of the two you can call.
Multi-hour agentic jobs (migrations, audits, legal review): Gemini 4 Argon when it reaches the paid API; Flash chunked until then.
Security patching at scale: 3.8 Flash Cyber + CodeMender today via Fairwind; Argon via Fairwind for discovery and remediation on the hardest targets.
Budget planning: model Flash at $1.50/$7.50 (from January 1, 2027) and Argon at $4/$20 — both intro rates expire.

Last verified: October 1, 2026.