Gemini 4 Argon costs 2.7 times as much per task as GPT-6.1 Sol at the same token price
Google's Gemini 4 Argon matches GPT-6.1 Sol's $2/$10 token price but costs 2.7 times as much per task, according to Artificial Analysis. Argon uses more tokens per job, so buyers still have to compare frontier models by cost per completed task.
The Scientist · Science desk
What happened
- The $2/$10 rate is a 50% launch discount on standard pricing of $4/$20, and Google has not confirmed when the promotion ends.
- On AA-Omniscience, Argon's hallucination rate is 15%, against 51% for GPT-6 Astra and 54% for GPT-6.1 Sol.
- Access is limited to more than 650 cyber defenders in Google's Fairwind Program, for defensive use only.
Why it matters
- What it costs cost Budgets set at the launch rate will understate spend, because Argon's per-task cost doubles whenever Google ends the promotion.
- Who has to decide decision Choosing between Argon and Sol has to come down to cost per completed task on the buyer's own workload, and on the public index Sol is far cheaper by that measure.
- What's now possible capability Workloads where a confident wrong answer is expensive now have a frontier option that declines to guess far more often than OpenAI's current models do.
- What it limits constraint Most enterprises cannot yet test Argon in production, so the only independent per-task cost evidence available to them comes from a single benchmarking firm.
Next story · Build · 6h ago
Gemini 4 Argon falls from 97.5% to 49.8% on ARC-AGI-3 once vendor features come offConfirmed2 publishers
Token rates are one input to a bill. The other is how much the model writes, and Artificial Analysis found Argon wordy. Across the full Intelligence Index, Argon generated 110 million tokens, against a median of 82 million for reasoning models in its price tier [11]. Per task, it writes about 2.3 times as many output tokens as GPT-6 Astra [20].
If the two models really charge the same rates, the gap with Sol comes from how many tokens each one uses. Argon costs $1.99 a task, 2.7 times Sol's figure, which puts Sol at about 74 cents a task [5][19]. That fits the firm's earlier finding that GPT-6.1 Sol runs at less than a quarter of GPT-6 Astra's cost per task. A quarter of Astra's $3.26 is about 82 cents [13][22]. Against Astra, Argon comes in at 60% of the per-task cost [5].
Forkast reported that Sol launched at the same rates a day before Argon [12], and it called the match "the commodity floor arriving in the frontier model tier" [26]. That floor holds only while Google's promotion runs. When it ends, Artificial Analysis expects Argon to cost $3.98 a task [15]. If OpenAI keeps Sol's price where it is, Argon would then cost about 5.4 times as much per task as Sol: the 2.7 multiple, doubled [23].
The evidence on what the extra money buys is mixed. Argon's headline coding score, 77.9% on DeepSWE v1.1, is Google's own figure, and Forkast notes that Google has not published independent verification [18]. The independent index behind the cost figures combines reasoning, knowledge, mathematics and coding [14].
The hallucination result is the most interesting number in the release, and it comes with lower accuracy. Argon gets 50% of AA-Omniscience questions right, 13 points below GPT-6 Astra. Its overall score of 42 sits beside Astra's 43 and Sol's 42 [17]. So Argon answers fewer questions correctly, but it guesses wrong far less often [16][17]. The thing this doesn't tell you is whether your workload is better served by a model that declines to answer.
The output limit is now 1 million tokens, up from 64,000 [9]. That changes what a single call can cost. A Gemini API feature called Long Decode Continuation pauses long responses and resumes them across follow-up calls, so reasoning can run that long without hitting request timeouts [10]. One full million-token response costs $10 in output tokens at the launch rate and $20 at standard pricing [21]. Even the context window is in dispute: Forkast reports 2 million tokens [24], while Artificial Analysis lists 1 million [25].
On a wider release, Google wrote that it will "continue to gather feedback from early testers as we iterate on guardrails before making Argon available to developers, enterprises, and consumers as soon as possible" [8].
What to watch
- Google confirming an end date for the 50% promotion, which moves Argon's per-task cost from $1.99 to $3.98.
- Independent DeepSWE v1.1 results for Argon against Claude Opus 5.5 and GPT-6 Astra, replacing Google's self-reported 77.9%.
- Any change to Sol's rates, or to Argon's token use at general release, since either would change the 2.7x per-task gap.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence68
- Adoption15
- Hype gap+20
- Incentives55
- Confidence65
Perspective Coverage
4 publishers- Builder
- Builder 36%
- Operator
- Operator 34%
- Investor
- Investor 30%
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Gemini 4 Argon is currently being rolled out to selected users and is not publicly available.
- [2]
Gemini 4 Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input token price.
- [3]
Google gated Argon behind its Fairwind Program, providing access to over 650 trusted cyber defenders, for defensive use only.
- [4]
Argon's standard pricing is $4/$20 per 1M input/output tokens, currently discounted 50% to $2/$10 for at least one month; Google has not yet confirmed the promotion end date.
- [5]
At current discounted pricing, Gemini 4 Argon (high) costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra (max, $3.26), but 2.7x GPT-6.1 Sol (max).
- [6]
Argon's cost efficiency versus GPT-6 Astra is driven by lower token prices rather than reduced token use; Argon averages 62k output tokens per task compared with 27k for GPT-6 Astra (max).
- [7]
Gemini 4 Argon (high) scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra (max, 53) and 1 point ahead of GPT-6.1 Sol (max, 52).
- [8]
We'll continue to gather feedback from early testers as we iterate on guardrails before making Argon available to developers, enterprises, and consumers as soon as possible.
- [9]
Google is expanding Argon's output token limit to 1M tokens, up from the previous 64K tokens.
- [10]
Long Decode Continuation is a Gemini API feature that pauses long responses and resumes them across follow-up calls, letting reasoning run up to 1M output tokens without request timeouts.
- [11]
When evaluated on the Intelligence Index, Argon (high) generated 110M tokens, compared with a median of 82M for reasoning models in a similar price tier.
- [12]
Twenty-four hours before Argon's September 30 announcement, OpenAI launched GPT-6.1 Sol at exactly the same rates ($2/$10 per million tokens).
- [13]
GPT-6.1 Sol scores 1 point below GPT-6 Astra on the Intelligence Index at less than one quarter of the Cost per Task.
- [14]
The Artificial Analysis Intelligence Index is a composite benchmark evaluating models across reasoning, knowledge, mathematics, and coding.
- [15]
After the discount ends, Argon's cost per Intelligence Index task will rise to $3.98 (about 1.2x GPT-6 Astra (max)).
- [16]
On AA-Omniscience, Gemini 4 Argon has a 15% hallucination rate, compared with 51% for GPT-6 Astra (max) and 54% for GPT-6.1 Sol (max).
- [17]
On AA-Omniscience accuracy, Argon scores 50%, 13 points below GPT-6 Astra (max, 63%); its overall AA-Omniscience score of 42 is in line with GPT-6 Astra (43) and GPT-6.1 Sol (42).
- [18]
Google reports 77.9% on DeepSWE v1.1; these are vendor-reported figures and Google has not published independent verification.
- [19]
GPT-6.1 Sol's implied cost per Intelligence Index task is about $0.74.
- [20]
Argon writes about 2.3 times as many output tokens per task as GPT-6 Astra.
- [21]
A full 1M-token output trajectory costs $10 in output tokens at the introductory rate and $20 at standard pricing.
- [22]
Less than a quarter of GPT-6 Astra's $3.26 per task is under about $0.82, consistent with Sol's implied $0.74.
- [23]
After the discount ends, Argon would cost about 5.4 times as much per task as GPT-6.1 Sol, assuming Sol's pricing is unchanged.
- [24]
Forkast reports Argon has a 2 million token context window.
- [25]
Artificial Analysis lists Argon's context window as 1M tokens.
- [26]
It is the commodity floor arriving in the frontier model tier.
Sources
4 independent publishers whose own reporting we read for this story.
- artificialanalysis.aiArtificial Analysis
2 articles · September 29, 2026
- deepmind.googleGemini 4 Argon: our next era of frontier intelligence
1 article · September 30, 2026
- forkast.newsGoogle’s Gemini 4 Argon Closes the Pricing Triangle. The Benchmark Lead Is the Real Story. – Forkast
1 article · September 30, 2026
- thenextweb.comGemini 4 Argon matches GPT-6 Astra and hallucinates far less in early tests
1 article · October 2, 2026