Models · scorecard

Gemini 4 Argon

What the vendor claims, next to what independent boards measured. The two are never mixed in one column.

At a glance

Maker: Google
Price per 1M tokens: $2 in · $10 out
Context window: 1.05M
Input → output: audio, file, image, text, video → text
Added to OpenRouter: Oct 13, 2026

Source: OpenRouter, as of Oct 28, 09:19 UTC. Model page on OpenRouter

What the vendor claims

Quoted word for word from the vendor's own launch post or model card. Claims, not measurements.

Scores the vendor claims, each with its source and the quoted line
Benchmark	Claimed	Source
Gray Swan Indirect Prompt Injection attack successSetting: 15 adaptive attemptsThe source says: Through built-in adversarial training and real-time security checks, Argon leads in prompt injection robustness on Gray Swan’s Indirect Prompt Injection (IPI) benchmark, cutting attack success across 15 adaptive attempts to 0.7%.	0.7%	Google DeepMind · Oct 13 From Drop #023 · Wed, Oct 14 · Morning
Shortcut or cheating attempts on coding benchmarksThe source says: On coding benchmarks where models have no incentive to hold back, attempts to take a shortcut or cheat on the task dropped to just 2% on Gemini 4 Argon, our best performance for recent Gemini models.	2%	Google DeepMind · Oct 13 From Drop #023 · Wed, Oct 14 · Morning

Independent results

Measured by independent sources (Epoch AI, LMArena, OpenRouter's benchmark runs, τ²-bench, Vectara), not by the vendor. "Listed as" is the source's own name for the model, so you can check our match.

Independent benchmark results for this model, each with its source and date
Board	Result	Rank
Epoch AI · GPQA diamondListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	96.4%Accuracy	1
Epoch AI · FrontierMath-Tiers-1-3-v2-PrivateListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	86.0%Accuracy	11
Epoch AI · FrontierMath-Tier-4-v2-PrivateListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	78.0%Accuracy	16
Epoch AI · SimpleQA VerifiedListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	76.0%Accuracy	1
Epoch AI · OTIS Mock AIME 2024-2025Listed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	100.0%Accuracy	1
Epoch AI · MirrorCodeListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	58.9%Accuracy	4
Epoch AI · Chess PuzzlesListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	59.0%Accuracy	10
Epoch AI · Mystery Game PuzzlesListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	64.0%Accuracy	6
Epoch AI · Furniture AssemblyListed as: gemini-4-argon_highSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	72.5%Accuracy	6
Epoch AI · Epoch Capabilities IndexListed as: Gemini 4 ArgonSource: Epoch AI (CC BY), as of Oct 28, 2026. Source	166.45Index	5
LMArena · Text (style control)Listed as: Gemini 4 Argon (High)LMArena (CC BY 4.0), as of Oct 22, 2026. Source	1,517Arena score	1
LMArena · WebDevListed as: Gemini 4 Argon (High)LMArena (CC BY 4.0), as of Oct 27, 2026. Source	1,680Arena score	10
LMArena · AgentListed as: Gemini 4 Argon (High)LMArena (CC BY 4.0), as of Oct 23, 2026. Source	0.117Score	6
OpenRouter · GPQA DiamondListed as: Gemini 4 Argonvia Google, its own providerSource: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Oct 28, 2026. Source	96.5%Accuracy	1
OpenRouter · GPQA Diamond: cost per taskListed as: Gemini 4 Argonvia Google, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Oct 28, 2026. Source	$0.095Cost per task	37
OpenRouter · GPQA Diamond: time per taskListed as: Gemini 4 Argonvia Google, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Oct 28, 2026. Source	102 sTime per task	32
OpenRouter · τ²-bench airlineListed as: Gemini 4 Argonvia Google, its own providerSource: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Oct 28, 2026. Source	78.7%Accuracy	8
OpenRouter · τ²-bench airline: cost per taskListed as: Gemini 4 Argonvia Google, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Oct 28, 2026. Source	$0.50Cost per task	53
OpenRouter · τ²-bench airline: time per taskListed as: Gemini 4 Argonvia Google, its own providerLower is better on this board.Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0, as of Oct 28, 2026. Source	348 sTime per task	41
Vectara · Hallucination rate in summariesListed as: google/gemini-4-argonLower is better on this board.Vectara Hallucination Leaderboard (HHEM-2.3), Apache-2.0, as of Oct 22, 2026. Source	12.1%Percent	83

Price history

OpenRouter's list price per 1M tokens in each of our snapshots.

Output now $10
Input now $2

Unchanged across our 29 snapshots.

Show the numbers

This model's OpenRouter price in each of our snapshots
Snapshot	Input $/1M	Output $/1M
Oct 28, 09:22 UTC	$2	$10
Oct 27, 16:03 UTC	$2	$10
Oct 27, 04:03 UTC	$2	$10
Oct 26, 16:04 UTC	$2	$10
Oct 26, 04:03 UTC	$2	$10
Oct 25, 16:04 UTC	$2	$10
Oct 25, 06:03 UTC	$2	$10
Oct 24, 15:04 UTC	$2	$10
Oct 24, 03:03 UTC	$2	$10
Oct 23, 15:03 UTC	$2	$10
Oct 23, 03:04 UTC	$2	$10
Oct 22, 15:03 UTC	$2	$10
Oct 22, 05:04 UTC	$2	$10
Oct 21, 15:03 UTC	$2	$10
Oct 21, 03:02 UTC	$2	$10
Oct 20, 15:04 UTC	$2	$10
Oct 20, 03:03 UTC	$2	$10
Oct 19, 15:04 UTC	$2	$10
Oct 19, 10:50 UTC	$2	$10
Oct 18, 15:03 UTC	$2	$10
Oct 18, 03:05 UTC	$2	$10
Oct 17, 15:03 UTC	$2	$10
Oct 17, 03:03 UTC	$2	$10
Oct 16, 15:02 UTC	$2	$10
Oct 16, 03:03 UTC	$2	$10
Oct 15, 17:02 UTC	$2	$10
Oct 15, 03:02 UTC	$2	$10
Oct 14, 15:03 UTC	$2	$10
Oct 14, 03:02 UTC	$2	$10

Source: OpenRouter, as of Oct 28, 09:22 UTC

ValueOur calculation

Capability vs price

This model in colour, against every other model on the Epoch Capabilities Index that OpenRouter prices.

1 model on the index has no OpenRouter price and is not plotted.

Show the numbers

Each model's capability index and sample-task price
Model	Epoch Capabilities Index	Sample task	Best value
Claude Opus 5.5Anthropic	168.83	$0.08	Yes
GPT-6 AstraOpenAI	168.81	$0.20
GPT-6.1 SolOpenAI	168.37	$0.04	Yes
Claude Fable 5.1Anthropic	166.49	$0.20
Gemini 4 ArgonGoogle	166.45	$0.04
Claude Sonnet 5.5Anthropic	166.18	$0.04
Claude Opus 5Anthropic	163.94	$0.10
GPT-6 SolOpenAI	163.73	$0.04
Claude Fable 5Anthropic	163.42	$0.20
GPT-5.5 ProOpenAI	162.91	$0.66
GPT-5.6 SolOpenAI	162.79	$0.08
Muse WonderMeta	161.85	$0.08
GPT-5.6 TerraOpenAI	160.09	$0.044
GPT-5.5OpenAI	159.68	$0.11
GPT-5.4 ProOpenAI	159.57	$0.66
Claude Haiku 5.5Anthropic	159.32	$0.002	Yes
Claude Opus 4.8Anthropic	159.01	$0.10
Gemini 3.7 FlashGoogle	157.97	$0.015
Kimi K3MoonshotAI	157.86	$0.042
Gemini 3.8 FlashGoogle	157.44	$0.017
GPT-5.4OpenAI	157.31	$0.055
Muse Spark 1.3Meta	157.21	$0.021
GPT-5.3-CodexOpenAI	157.09	$0.045
Claude Sonnet 5Anthropic	156.88	$0.04
Claude Opus 4.7Anthropic	156.79	$0.10
GPT-6 LunaOpenAI	156.75	$0.002
Grok 4.6xAI	156.57	$0.032
GPT-5.6 LunaOpenAI	156.5	$0.0044
GPT-5.2 ProOpenAI	156.33	$0.55

Our calculation: Epoch AI's index against the sample-task price (undefined tokens in, undefined out, at each model's OpenRouter list price; the formula is under the price chart).

Source: Epoch AI (CC BY), as of Oct 28, 2026Source

Source: OpenRouter, as of Oct 28, 09:19 UTC

In the Drops

Models

Google DeepMind says Gemini 4 Argon cuts prompt-injection success to 0.7%

In a safety post, Google DeepMind says Argon solved all 12 simulated cyberattack scenarios, up from five, so it added input screening, activation probes and live monitoring. It says shortcut attempts fell to 2% on coding tests.

Why it matters: For agent builders, Google says a monitor reads Argon's actions and chain of thought in real time, and the model was trained to pause for human confirmation.

Google DeepMind · Oct 13 Hacker News · Oct 13

From Drop #024 · Wed, Oct 14 · Evening

Models

Gemini 4 Argon tops Epoch AI's SimpleQA Verified and LMArena's text board

Epoch AI lists Gemini 4 Argon (high) first on SimpleQA Verified at 0.76, ahead of GPT-6 Astra at 0.756. LMArena's text board with style control ranks it first at 1524.768 as of 2026-10-13.

Why it matters: Two independent boards put Argon on top for factual recall and chat preference, a day after launch.

Epoch AI (CC BY) · Oct 13 LMArena (CC BY 4.0) · Oct 13

From Drop #024 · Wed, Oct 14 · Evening

Models

Gemini 4 Argon tops Epoch AI's GPQA diamond at 0.964

Epoch AI lists gemini-4-argon_high first on GPQA diamond at 0.964 and on SimpleQA Verified at 0.76. LMArena's text board as of 2026-10-08 ranks gemini-4-argon-high first at 1525.396.

Why it matters: Independent boards put Argon on top for knowledge tests. On LMArena's agent board as of 2026-10-12, it ranks fourth at 0.124.

Epoch AI (CC BY) · Oct 13 LMArena (CC BY 4.0) · Oct 12

From Drop #023 · Wed, Oct 14 · Morning

Models

Gemini 4 Argon leads LMArena's text board as of October 8

On LMArena's text board with style control, as of 2026-10-08, gemini-4-argon-high scores 1525.396, ahead of claude-opus-5.5-high at 1507.076 and claude-opus-4-6-high at 1504.355.

Why it matters: Leaders differ by task: on LMArena's WebDev board as of the same date, gemini-4-argon-high is ninth at 1677.806, behind claude-opus-5.5-max at 1813.028.

LMArena (CC BY 4.0) · Oct 8

From Drop #017 · Sun, Oct 11 · Morning

Models

Gemini 4 Argon's $2/$10 price is introductory; Google says it will double

CNBC reports Argon's intro price of $2/$10 per million tokens matches GPT-6.1 Sol, but Google says rates will rise to $4/$20. Access is still limited to select cyber defenders and cloud customers, with no public date.

Why it matters: If you are budgeting for Argon, plan for the doubled rate. Broad access, and with it independent testing, still has no date.

CNBC · Oct 1 CNBC · Oct 2

From Drop #002 · Fri, Oct 2 · Evening