DEV Community

Cover image for Gemini 4 Argon Cost per Task, Measured: $1.99 Now, $3.98 at Standard Prices
AIHubMix
AIHubMix

Posted on

Gemini 4 Argon Cost per Task, Measured: $1.99 Now, $3.98 at Standard Prices

#ai #api #gemini #llm

On Artificial Analysis' Intelligence Index, one Gemini 4 Argon task costs $1.99 at the launch price and $3.98 at the standard price. On the Vals Index, run at standard prices, it costs $15.68 per test, rising to $193.78 on a long computer-use benchmark.

Those are the numbers to size budgets, spend caps and rate limits around, not the $2/$10 headline. The engineering problem is that you have to plan before you can call the model, with part of the price list still unpublished. This post lays out the measured per-task costs, explains why the per-token discount doesn't carry through, walks through the billing failure modes that break a budget, each with a fix, and ends with a small calculator for your own token counts. Inputs Google hasn't published are marked TBD.

The price inputs

Argon's prices come from Google's launch post; the other Gemini rows come from the Gemini API pricing page. USD per million tokens.

Model Input Cached input Output
Gemini 4 Argon, promo $2.00 $0.10 $10.00
Gemini 4 Argon, standard $4.00 TBD $20.00
Gemini 3.1 Pro Preview $2.00 $0.20 $12.00
Gemini 3.8 Flash, to Dec 31 $0.75 $0.075 $3.75
Gemini 3.8 Flash, from Jan 1 $1.50 $0.15 $7.50
GPT-6 Astra $10.00 — $50.00
  • Gemini 3.1 Pro Preview prices apply to prompts up to 200k tokens; above that, input is $4 and output $18.
  • GPT-6 Astra's price is as reported by Engadget.
  • Google hasn't said when the promotion ends. Artificial Analysis says it runs for at least one month from September 30, 2026.
  • The 95% cache discount may or may not outlast the promotion. For comparison, Gemini 3.8 Flash's cache discount is 90%.

Measured cost per task

Figures in this section come from Artificial Analysis (its launch analysis and model pages) and Vals AI. Both ran Argon at high, its highest reasoning setting. Neither is Google's own number.

Artificial Analysis: promo and standard prices

Artificial Analysis runs every model through the same Intelligence Index tasks and reports the average cost of one task. It doesn't say how many tasks the index has.

Model Index score Cost per task
Gemini 4 Argon (high), promo 53 $1.99
Gemini 4 Argon (high), standard 53 $3.98
GPT-6 Astra (max) 53 $3.26
Gemini 3.1 Pro Preview 30 $1.30
  • Argon's promo token price is a fifth of Astra's; its cost per task is three fifths. An 80% discount per token shrinks to 40% per task, and at standard prices Argon costs about 1.2 times as much as Astra.
  • The reason is token count. Argon averaged about 62k output tokens per task against about 27k for Astra, roughly 2.3 times as many. Running the whole index it generated 110M output tokens; the median model generated 85M, Gemini 3.1 Pro Preview 67M. Artificial Analysis puts it this way: the saving "is driven by lower token prices, rather than reduced token use".
  • Against Gemini 3.1 Pro Preview, Argon scores 23 points higher and costs about 1.5 times as much per task at promo prices, about 3 times at standard prices.

Vals AI: standard prices only

Vals AI ran Argon at the standard $4/$20 price, with temperature 1 and a 262,144-token output cap, and reports the average cost of one test.

Model Vals Index Cost per test
Gemini 4 Argon 68.90% $15.68
Claude Sonnet 5.5 67.04% $21.34
Claude Opus 5.5 66.97% $32.14
Claude Fable 5.1 65.83% $28.71

Even at standard prices, Argon costs less per test on this suite than the three Claude models just below it. The cost climbs on Vals' long agentic benchmarks:

Vals benchmark Argon score Cost per test
CUA-bench 4.83% $193.78
Code Migration 68.17% $57.82
Terminal-Bench Science 44.29% $44.90
SRE Bench 44.28% $30.45

CUA-bench, a computer-use benchmark, is the worst case: the most expensive test in the set and one of Argon's weakest results, 7th of the 8 models Vals ran.

Reading the two sets together

  • Against GPT-6 Astra, the promotion carries the saving. Against Claude, Argon is cheaper per test even at standard prices.
  • Long agentic runs are where the bill grows: $30 to $194 per test on Vals' agentic benchmarks, against $15.68 on the Vals Index.
  • Compare models within one tester, not across them. Artificial Analysis quotes the promo price with the standard figure alongside; Vals uses the standard price. Their tasks and harnesses differ too.

Failure modes and fixes

The monthly cap trips in half the expected time

Cause: the budget was set at $2/$10, and the promotion has no announced end date. When it ends, the same traffic spends twice as fast.

Fix: budget at $4/$20 from day one and treat the promotion as a bonus.

One request costs $10

Cause: Argon's output limit is 1M tokens. An uncapped answer that runs to the limit costs $10 at promo prices and $20 at standard.

Fix: set an output cap on every request, sized to the job, instead of leaving the 1M ceiling open.

The output line is much bigger than the visible answers

Cause: Argon writes long: about 62k output tokens per Artificial Analysis task, 2.3 times GPT-6 Astra's 27k. On Gemini 3 models, Google also bills output "including thinking tokens". Argon isn't on Google's pricing page yet, so that's worth confirming for Argon.

Fix: fix the thinking level per workflow, and budget output tokens for the reasoning, not just the answer.

Cached input gets billed at full price

Cause: the AIHubMix Gemini guide notes that an implicit cache hit needs the same content, model and parameters, with repeated context placed first. Changing the thinking level mid-session or reordering the prompt between turns turns cached input back into full-price input.

Fix: put stable context first and keep it identical from turn to turn: system prompt, tool definitions and documents at the top, the user's new message last. Don't switch thinking levels mid-session.

Declined answers get paid for twice

Cause: Argon declines more often than other frontier models when it isn't sure. A declined answer still bills its input and thinking tokens, and retrying the same question on Argon pays again.

Fix: route declines to a person or another model rather than retrying Argon.

Long answers turn into several billed calls

Cause: Long Decode Continuation pauses long answers and resumes them across follow-up calls. How those calls are billed, and whether earlier output is counted again as input, isn't documented. [TBD: Long Decode Continuation billing]

Fix: none confirmed yet. The per-request output cap still bounds each call.

Easy steps billed at frontier rates

Cause: every step of the agent runs on Argon, including extraction and formatting.

Fix: route by difficulty. Send extraction, classification and formatting to Gemini 3.8 Flash and keep Argon for the hard reasoning steps. Gemini 3 models also get 50% off through the Batch API; Argon's batch rate is TBD, so re-check when it appears.

Estimate your own cost

Prices go in a dict, and the standard cached rate is an argument because nobody has published it. Plain Python, no dependencies. Feed it token counts you measured yourself.

PRICES = {  # USD per 1M tokens
    "promo": {"input": 2.00, "cached": 0.10, "output": 10.00},
    "standard": {"input": 4.00, "cached": None, "output": 20.00},  # cached rate TBD
}


def task_cost(uncached_in, cached_in, output, tier="promo", cached_rate=None):
    p = PRICES[tier]
    cached = p["cached"] if p["cached"] is not None else cached_rate
    if cached is None:
        raise ValueError("Pass cached_rate: Argon's standard cached price isn't published yet")
    return (uncached_in * p["input"] + cached_in * cached + output * p["output"]) / 1_000_000


# Use token counts from your own logs, e.g. the usage field of the same workflow
# run on Gemini 3.8 Flash: task_cost(uncached_in, cached_in, output, "promo")
Enter fullscreen mode Exit fullscreen mode

Measure real token counts on Gemini 3.8 Flash or 3.1 Pro Preview through AIHubMix and feed them into task_cost. Expect Argon's to be higher: on Artificial Analysis' index it generated 110M output tokens where Gemini 3.1 Pro Preview generated 67M. Treat the result as a floor, not a forecast.

What you can't pin down yet

  • When the promotion ends
  • The cached-input rate at standard pricing, any long-context surcharge, Batch and Flex rates, and cache storage. [TBD: Argon standard cache rate, long-context tier, Batch/Flex rates]
  • Long Decode Continuation billing (above)
  • Whether thinking tokens are billed as output on Argon, as they are on Gemini 3
  • Argon's token counts on your own workload

Getting ready on AIHubMix

Argon has a model page on AIHubMix, but it's marked coming soon and its AIHubMix price isn't set yet. [TBD: AIHubMix price for Gemini 4 Argon]

Until then, Gemini 3.8 Flash is the cheapest way to prototype the same workflow and collect token counts. Argon's price will appear on its model page once it's set.

Keep reading: the Gemini 4 Argon series

Sources

Top comments (0)