Skip to content

Gemini 4 Argon vs Claude Opus 5.5 vs Sonnet 5.5: Which to Use

Argon is $2/$10 per million tokens through November 30, 2026, then $4/$20. Here is where you can run it on your plan and what each Claude tier costs next to it.

•••20 min read•AI Model Comparison
Step chart: Gemini 4 Argon's price sits at Claude Sonnet 5.5's $2 / $10 level from October 13 and steps up to Claude Opus 5.5's $4 / $20 on December 1

Gemini 4 Argon is now callable. Google made it generally available on October 13, 2026, as gemini-4-argon on the Gemini API paid tier, in the Gemini app and AI Mode in Search for Google AI Pro and Google AI Ultra subscribers, and in GitHub Copilot for Pro+, Max, Business, and Enterprise. There is no free tier. Its price is the part that decides the comparison: through November 30, 2026, Argon costs exactly what Claude Sonnet 5.5 costs per token, and from December 1 it costs exactly what Claude Opus 5.5 costs.

  • If you run Claude Sonnet 5.5 to keep costs down: trial Argon now, while both cost $2 input / $10 output per million tokens. From December 1, Argon costs twice as much as Sonnet 5.5 for the same tokens, so it has to win on your tasks by enough to justify that.
  • If you run Claude Opus 5.5 for hard, open-ended work: from December 1, Argon is the same price, so the question is quality on your own tasks. Google's own table shows Opus 5.5 ahead on terminal and long software-engineering benchmarks; Argon's largest claimed leads are in legal, finance, science, and long-context work.
  • If you are choosing between the two Claude models: Sonnet 5.5 for well-scoped tasks with a clear pass/fail check, Opus 5.5 for long, open-ended jobs. That advice has not changed.

The two independent indexes that score all three models disagree on the order, so run your own task set on whichever Argon channel you already have before moving a workload.

Gemini 4 ArgonClaude Opus 5.5Claude Sonnet 5.5
Where developers can call itGemini API paid tier, GitHub Copilot (Pro+, Max, Business, Enterprise)Claude API, Amazon Bedrock, Google Cloud, Microsoft FoundrySame as Opus 5.5
API model IDgemini-4-argonclaude-opus-5-5claude-sonnet-5-5
Input / output per 1M tokens$2 / $10 through November 30, 2026; $4 / $20 from December 1$4 / $20$2 / $10
Cached input per 1M tokens$0.10, then $0.20 from December 1 (implicit caching only)$0.20 cache read$0.10 cache read
Context window1,048,576 tokens, shared by input and output1M tokens1M tokens
Maximum output65,536 by default; up to 1,048,576 with long decode128K tokens128K tokens
Vals Index v2.1 (October 13, 2026)68.90%66.97%67.04%
Artificial Analysis Intelligence Index v4.3.253 (High)58 (Max)56 (Max)

Prices are USD list prices on each vendor's own API, before tax.

Where you can use Gemini 4 Argon since October 13, plan by plan

Google's launch post says Argon is "globally available today for developers and enterprises via the API in AI Studio and Gemini platform, for Google AI Ultra subscribers in Google Antigravity, and Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search." The Gemini API changelog lists it as generally available on October 13, and the model page marks gemini-4-argon as Stable. Before that date, after its September 30 announcement, only cyber defenders in Google's Fairwind Program and trusted testers had access.

WhereWho gets ArgonHow you pay
Gemini API (AI Studio keys, Android Studio)Any paid-tier project; the Free tier is "Not available"Per token, at the prices in the next section
Gemini appGoogle AI Pro ($19.99/month) and Google AI Ultra (from $99.99/month); not Free or Google AI PlusSubscription
AI Mode in SearchGoogle AI Pro ("higher access") and Google AI Ultra ("highest access")Subscription
AI Studio playground on a Google AI planGoogle AI Ultra only; "not available on the Free or Google AI Pro plans"Subscription quota, then a paid API key
Google AntigravityGoogle AI Ultra subscribers; enterprisesSubscription
GitHub CopilotCopilot Pro+, Max, Business, and Enterprise; not Copilot Free or Copilot ProUsage-based billing "at provider list pricing"
Gemini for businessEnterprise customersGemini for business plan

US list prices come from Google's Gemini subscriptions page. The free Gemini app runs Flash-Lite, and Google AI Plus ($4.99/month) gets the Flash model, not Argon.

The "not for Pro users" confusion comes from two different things that share the word Pro. The Gemini API pricing page says Argon "is not available on the Free tier or on the Google AI Pro plan." That refers to the AI Studio subscription: Google's AI Studio plans page says plan benefits apply only inside the AI Studio web interface, and Argon there is Ultra-only. A Google AI Pro subscriber still gets Argon in the Gemini app and AI Mode, and anyone can call it through the API with a paid key, which is billed separately from any subscription.

In GitHub Copilot, the changelog entry puts Argon in the model picker in VS Code, Visual Studio, Copilot CLI, the Copilot coding agent, the GitHub Copilot app, JetBrains IDEs, Xcode, and Eclipse. The rollout is gradual, and Business and Enterprise admins control it through model policy (new models are on by default). GitHub does not say whether "provider list pricing" means the $2 / $10 introductory rate or the $4 / $20 rate.

Two narrow no-cost routes exist, both by application: scientists and researchers can apply for six months of Antigravity's highest usage tier, subject to eligibility and capacity, and cyber defenders can apply to the Fairwind Program.

On the Claude side, Opus 5.5 (released September 22, 2026) and Sonnet 5.5 (released September 28, 2026) are generally available on the Claude API, Amazon Bedrock (anthropic.claude-opus-5-5, anthropic.claude-sonnet-5-5), Google Cloud, Microsoft Foundry, and Claude Platform on AWS, according to Anthropic's models overview. Whether a given cloud region has them enabled depends on your account. Both take text and images in and return text, with a 1M-token context window and a 128K synchronous output limit.

Argon's limit works differently. Its 1,048,576-token context window is a "unified input and output token limit," so a 900,000-token prompt leaves about 148K tokens for the answer. Default synchronous output is 65,536 tokens; reaching the full 1M output requires Google's long-decode continuation setup. Argon accepts text, images, video, audio, and PDF and returns text.

Argon starts at Sonnet 5.5's price, then moves to Opus 5.5's on December 1

Argon's price sheet maps onto Anthropic's almost exactly. The Gemini API pricing page lists $2 input / $10 output per million tokens "through November 30, 2026," which is Sonnet 5.5's list price, and $4 / $20 "starting December 1, 2026," which is Opus 5.5's. The launch post calls the first rate introductory.

Per 1M tokens (USD)Argon, through November 30Argon, from December 1Claude Opus 5.5Claude Sonnet 5.5
Input$2$4$4$2
Output (Argon: including thinking tokens)$10$20$20$10
Cached input / cache read$0.10$0.20$0.20$0.10
Cache write (5-minute)No charge listed (implicit only)No charge listed (implicit only)$5$2.50
Batch input / output$1 / $5$2 / $10$2 / $10$1 / $5

The Claude figures come from Anthropic's pricing page, where a cache hit on both 5.5 models costs 5% of input: $0.20 on Opus 5.5 and $0.10 on Sonnet 5.5. That puts the 2× gap between the two Claude models on every token type, cached input included. Opus 5.5 also has a fast mode in research preview at $8 / $40, Claude API only, and US-only inference multiplies every Claude rate by 1.1. Bedrock and Google Cloud set their own Claude prices.

Argon's cached input is 95% off input in both periods, which lines it up with Sonnet 5.5 before December 1 and with Opus 5.5 after. The difference is control. Argon supports implicit caching only, so Google decides when a repeated prefix counts as a hit. Claude's caching is explicit: you mark the prefix, pay a write fee when it is stored, and know which tokens will bill at the read rate. Argon also has a Priority tier at $3.60 / $18 through November 30 and $7.20 / $36 from December 1, and its Free tier is listed as "Not available" on every line. Grounding with Google Search on Argon includes 5,000 free requests a month, shared across Gemini 3 and newer models, then costs $14 per 1,000. These are Gemini API rates; Vertex AI on Google Cloud keeps its own price list.

Recompute the cost for your own task

The formula is the same for all three: tokens in millions times the per-million rate, summed across input, cached input, and output.

Example A: 100,000 input tokens and 20,000 output tokens, no caching.

ModelCalculationCost
Claude Sonnet 5.50.1 × $2 + 0.02 × $10$0.40
Claude Opus 5.50.1 × $4 + 0.02 × $20$0.80
Gemini 4 Argon, through November 300.1 × $2 + 0.02 × $10$0.40
Gemini 4 Argon, from December 10.1 × $4 + 0.02 × $20$0.80

Example B: one turn of a cache-heavy agent, with 200,000 cached input tokens, 20,000 fresh input tokens, and 5,000 output tokens.

ModelCalculationCost
Claude Sonnet 5.50.2 × $0.10 + 0.02 × $2 + 0.005 × $10$0.11
Claude Opus 5.50.2 × $0.20 + 0.02 × $4 + 0.005 × $20$0.22
Gemini 4 Argon, through November 300.2 × $0.10 + 0.02 × $2 + 0.005 × $10$0.11
Gemini 4 Argon, from December 10.2 × $0.20 + 0.02 × $4 + 0.005 × $20$0.22

With Sonnet 5.5's cache read at $0.10, Example B on Sonnet costs exactly half of Opus 5.5, the same 50% as Example A. Argon matches Sonnet to the cent until November 30 and Opus from December 1.

Example cache-heavy agent turn: $0.11 on Claude Sonnet 5.5, $0.22 on Claude Opus 5.5, Gemini 4 Argon at $0.11 through November 30 and $0.22 from December 1, and $0.49 or $0.98 if Argon's implicit cache misses

Example B leans on two assumptions that cut in opposite directions. Claude's first turn pays to write the 200,000-token prefix: $0.50 on Sonnet 5.5 and $1.00 on Opus 5.5 at the 5-minute rate (0.2 × $2.50 and 0.2 × $5), spread over every later turn that reads it. Argon has no write fee, but if its implicit cache does not hit, all 220,000 input tokens bill at the full rate: 0.22 × $2 + 0.005 × $10 = $0.49 through November 30, and 0.22 × $4 + 0.005 × $20 = $0.98 from December 1. Check how many input tokens your API responses report as cached to see which case you are in.

The December 1 change is easiest to see as a monthly bill. Run 10,000 Example A tasks a month and Sonnet 5.5 costs $4,000, Opus 5.5 costs $8,000, and Argon costs $4,000 in November and $8,000 in December, with nothing about your workload changing.

All of these examples assume the three models consume and produce the same number of tokens, which they will not. Tokenizers differ, and effort and thinking settings change how many tokens a model spends on the same task. Opus 5.5 defaults to medium effort and Sonnet 5.5 to high; Argon has low, medium, and high thinking levels, and its output price includes thinking tokens. To get your real cost per task, take the token counts from your own API responses and plug them into the formula. For the billing rules behind thinking tokens, see how reasoning tokens are billed on OpenAI, Claude, and Gemini.

Who measured each benchmark number

Several sets of numbers are in circulation, and each comes from a different party with a different setup. All figures below are published by a vendor or a leaderboard, and each is labeled with who ran it.

Google's table compares Argon with Opus 5.5 only

Google's model page puts Argon next to Opus 5.5 on 18 rows. Sonnet 5.5 is not in the table.

Benchmark (%)Gemini 4 ArgonClaude Opus 5.5
Vals Index68.967.0
AutomationBench51.342.5
Vals Finance Agent v265.458.6
Harvey Legal Agent Benchmark19.63.8
DeepSWE v1.177.974.2
Vibe Code Bench91.990.3
LABBench 288.873.1
RiemannBench76.069.6
GraphWalks, up to 128K99.790.6
GraphWalks, 256K–1M84.266.8
Agent's Last Exam39.538.2
Chartography71.666.3
LVBench91.783.7
CWE-bench v168.067.0
FrontierSWE v255.062.3
Terminal-Bench 4.057.466.4
PostTrainBench45.349.3
Terminal-Bench Science 0.157.663.3

Argon is ahead on 14 rows and Opus 5.5 on the last four, which cover software engineering, terminal work, and ML engineering. What the table can support is narrower than the row count suggests. Google chose the benchmarks. Its evaluation methodology says Argon ran at the highest thinking setting, while most Opus 5.5 figures are Anthropic's self-reported numbers or public leaderboard entries at maximum or best-available reasoning. Google computed several rows itself for every model: PostTrainBench, LABBench 2, GraphWalks, and LVBench. On LVBench the models did not even see the same input, with Gemini sampled at one frame per second and Opus 5.5 given 600 frames "due to API limitations."

Read it as a list of where to look, not as a ranking. The largest gaps (legal agent tasks, long-context graph traversal, lab-bench science) are the ones worth reproducing on your own data. The rows Opus 5.5 wins tell you where Argon is least likely to displace it.

Vals Index puts Argon first; Artificial Analysis puts it last

Two independent leaderboards score all three models, and they put them in opposite order.

The Vals Index is run by Vals AI, not by either vendor. It weights finance, coding, legal, and tax tasks by their share of GDP. Version 2.1, as of its October 13, 2026 update, shows:

ModelVals Index scoreCost per testTime per test
Gemini 4 Argon68.90%$15.6846m 33s
Claude Sonnet 5.567.04%$21.341h 18m
Claude Opus 5.566.97%$32.141h 19m

The Artificial Analysis Intelligence Index v4.3.2 runs each model at a stated setting and reports the cost of running the whole index:

Model (setting)Intelligence IndexCost to run the index
Claude Opus 5.5 (Max)58$8,708
Claude Sonnet 5.5 (Max)56$7,260
Gemini 4 Argon (High)53$2,407

The two scales measure different things. Vals scores agent work on economic tasks, and there Argon leads by just under two points while finishing in roughly 60% of the time. Artificial Analysis combines its own evaluation suite into one score, and there both Claude models lead, with Argon run at High against Claude at Max. Both agree on one thing: Argon is the cheapest of the three to run on their workloads, and Sonnet 5.5 sits between Argon and Opus 5.5 on cost. Artificial Analysis lists Argon at $2 / $10, so its $2,407 reflects the introductory rate; doubled for the December 1 price it would be about $4,814, still below both Claude models. On Vals, Sonnet 5.5 and Opus 5.5 are 0.07 points apart, which Vals describes as effectively tied, at about two-thirds of Opus's cost per test.

Vals does not say which Argon price its $15.68 assumes. If it is the $2 / $10 rate and token use stays the same, the December 1 price roughly doubles it to $31.36, level with Opus 5.5. If it already uses $4 / $20, Argon stays the cheapest per test after December 1. Leaderboard rows also move with reruns, so quote these scores with their date and treat a gap under two points as unsettled.

Vals Index v2.1 ranks Gemini 4 Argon first at 68.90% while the Artificial Analysis Intelligence Index v4.3.2 ranks it last at 53, with each model's cost per test and cost to run the index

Anthropic's numbers compare Sonnet 5.5 with Opus 5.5

Anthropic's Sonnet 5.5 announcement is the only place the two Claude models are measured side by side by their maker. Higher is better on every row.

BenchmarkClaude Sonnet 5.5Claude Opus 5.5
Terminal-Bench 4.070.6%66.4%
FrontierCode 1.146.2%54.4%
CursorBench 4.055.5%57.8%
GDPval-AA v2.118441846
AA-Briefcase v1.118111822
Humanity's Last Exam (with tools)64.5%67.7%
OSWorld 2.180.1%81.8%
Chartography (no tools)61.6%64.4%

Sonnet 5.5 leads on Terminal-Bench 4.0, where Opus's 66.4% is its best result (at xhigh effort). Opus 5.5 leads everywhere else, by about two to three percentage points on most rows and by eight on FrontierCode 1.1. That FrontierCode gap narrows when Sonnet is run at xhigh (52.1%) instead of max (46.2%). Anthropic's own summary matches the table: Sonnet 5.5 is for "well-scoped everyday tasks," while on complex, open-ended work Opus 5.5 "remains clearly stronger."

Two cautions on reading across tables. Anthropic lists Opus 5.5's Chartography at 64.4% and Google's table lists it at 66.3%. Those are different runs under different settings, so compare numbers only within one table. And Anthropic's claims that Sonnet 5.5 "generates outputs 30%+ faster" and "costs up to 30% less per task" are measured against Sonnet 5, not against Opus 5.5 or Gemini.

There is no vendor-published head-to-head between Argon and Sonnet 5.5. Vals and Artificial Analysis are the only places the two share a scale, and they disagree.

Sonnet 5.5 vs Opus 5.5: which Claude model to use now

The evidence above points to a simple rule: choose by how open-ended the task is, then confirm with your own token counts.

Your workloadStart withWhy
Code generation, data analysis, content, tool-using agents on bounded tasksClaude Sonnet 5.5Tied with Opus 5.5 on the Vals Index at about two-thirds of the cost per test; half the token price; "Fast" latency label versus "Moderate"
Multihour autonomous coding agents, large-scale refactoring, complex systems engineeringClaude Opus 5.5Anthropic says it "remains clearly stronger" on complex, open-ended work; eight-point lead on FrontierCode 1.1
Computer use and vision-heavy workflowsClaude Opus 5.5Anthropic's model selection guide assigns these to Opus; small leads on OSWorld 2.1 and Chartography
Latency-sensitive tool loopsClaude Sonnet 5.5It can skip up-front thinking with thinking: {"type": "between_tools"} at low, medium, or high effort; Opus 5.5's adaptive thinking is always on
Cache-heavy agents where most input is a cached prefixClaude Sonnet 5.5, unless quality failsCache reads cost $0.10 on Sonnet and $0.20 on Opus, so the saving stays at 50% (Example B)

Anthropic's selection guide says "most workloads start with Claude Opus 5.5," and that is a reasonable default if quality on hard tasks matters more to you than the bill. The cost-first path runs the other way. Start on Sonnet 5.5, keep a set of your own tasks with a pass/fail check, and move a workload to Opus 5.5 only when Sonnet fails that check. Before switching in either direction, try changing effort. Anthropic notes that tuning effort "is often a better lever than switching models," and the guide to choosing and setting effort on Opus 5.5 shows how to measure what each level costs.

If you are moving from an older Claude model, both 5.5 models reject forced tool use (tool_choice set to any or tool returns a 400) and neither accepts thinking: {"type": "disabled"}. The full list is in the Claude Opus 5.5 migration checks, and Sonnet 5.5's differences are in Anthropic's what's new page. For where Haiku and Fable fit, see which Claude model to use.

What to test on Argon before the price doubles on December 1

The introductory window is the cheap time to find out whether Argon earns a place in your stack. Run the same pass/fail task set you use for Claude on whichever Argon channel you already have: an API key on the paid tier, the Gemini app on Google AI Pro or Ultra, or the Copilot model picker on Pro+ and above. Only the API gives you exact token counts; in the app you can judge the answers but not the cost.

It is most worth a trial in three cases:

  • You run Sonnet 5.5 and price is the constraint. Through November 30 the bill is the same per token. Compare pass rate and tokens per task, not rate cards, and decide whether Argon's results justify paying twice Sonnet's rate from December 1.
  • You run Opus 5.5 for long-context, legal, finance, or science work. Those are the rows with Google's largest claimed leads, and from December 1 the two cost the same per token. Check them on your own documents.
  • You need very long outputs. Argon can produce up to 1,048,576 tokens with long decode, against 128K on both Claude models, though the output shares the same budget as the prompt.

It is less likely to pay off if your workload looks like the rows Opus 5.5 wins in Google's own table, or where Sonnet 5.5 posts its best result: terminal-driven agents and long software-engineering tasks.

Three things to check during the trial. First, how many of your input tokens bill as cached: Argon's caching is implicit only, and Example B shows a missed cache more than quadruples the cost of a cache-heavy turn. Second, whether Batch or Flex fits: both are half the standard rate, so after December 1 Argon's Batch rate ($2 / $10) equals Sonnet 5.5's standard rate and Opus 5.5's Batch rate. Third, if you use Copilot, which rate appears on your usage report, since GitHub only says "provider list pricing."

If you still need a cheaper Gemini model, Gemini 3.8 Flash costs $0.75 input / $3.75 output per million tokens through December 31, 2026, and has a free tier; the Gemini 3.8 Flash API guide covers its model ID, pricing, and limits. It is a different class of model from Argon, so treat it as a separate decision.

Questions about Gemini 4 Argon, Opus 5.5, and Sonnet 5.5

Is Gemini 4 Argon free?

No. The Gemini API lists Argon's Free tier as "Not available," the free Gemini app runs Flash-Lite, and Google AI Plus does not include it. You need a paid API key, a Google AI Pro or Ultra subscription (Gemini app and AI Mode), Google AI Ultra (AI Studio playground and Antigravity), or Copilot Pro+, Max, Business, or Enterprise with usage-based billing. The only no-cost routes are selective applications: six months of Antigravity for accepted scientists and researchers, and the Fairwind Program for cyber defenders. For Gemini models that do have a free API tier, see what the Gemini API free tier covers.

Where can I use Gemini 4 Argon?

As of October 14, 2026: the Gemini API paid tier (AI Studio keys and Android Studio), the Gemini app and AI Mode in Search on Google AI Pro or Ultra, the AI Studio playground and Google Antigravity on Google AI Ultra, GitHub Copilot on Pro+, Max, Business, and Enterprise, and Gemini for business. The Copilot rollout is gradual, so the model may appear in your picker a few days later.

Can I use Gemini 4 Argon through the API right now?

Yes, since October 13, 2026. Use the model ID gemini-4-argon from a project on the paid tier; the free tier does not include it. It costs $2 input / $10 output per million tokens through November 30, 2026, and $4 / $20 from December 1. A Google AI Pro or Ultra subscription does not cover API calls, which are billed separately.

Is Gemini 4 Argon cheaper than Claude Opus 5.5?

Through November 30, 2026, yes: $2 / $10 per million tokens is half of Opus 5.5's $4 / $20 and equal to Sonnet 5.5. From December 1, Argon costs $4 / $20, the same as Opus 5.5, with cached input at $0.20 on both. Equal token prices do not guarantee equal cost per task, because the models use different numbers of tokens on the same work.

Is Gemini 4 Argon better than Claude Opus 5.5?

In Google's own table it leads on 14 of 18 benchmarks and trails on FrontierSWE v2, Terminal-Bench 4.0, PostTrainBench, and Terminal-Bench Science 0.1. On the Vals Index it leads by about two points; on the Artificial Analysis Intelligence Index, Opus 5.5 leads 58 to 53. The Google table is vendor-selected and mixes self-computed and self-reported results, so it shows where Argon may be stronger, not by how much on your workload.

Is Gemini 4 Argon better than Claude Sonnet 5.5?

Neither vendor has published a head-to-head. On the Vals Index, Argon scores 68.90% to Sonnet 5.5's 67.04%; on the Artificial Analysis Intelligence Index, Sonnet 5.5 scores 56 to Argon's 53. The price is identical through November 30, 2026, and Argon costs twice as much from December 1, so the trial period is the time to compare them on your tasks.

Is Sonnet 5.5 better than Opus 5.5?

On one benchmark in Anthropic's table, Terminal-Bench 4.0, yes (70.6% versus 66.4%). On the Vals Index the two are effectively tied. On Anthropic's other seven published benchmarks Opus 5.5 is ahead, mostly by small margins. Sonnet 5.5 is the better value on well-scoped tasks; Opus 5.5 is the stronger model on complex, open-ended ones.

What is Gemini 4 Argon's context window?

1,048,576 tokens, shared by input and output. Google's model page calls it a "unified input and output token limit," so the longer your prompt, the less room is left for the answer. Default synchronous output is 65,536 tokens; long decode raises it to the full 1,048,576. Claude Opus 5.5 and Sonnet 5.5 have a 1M-token context window with a separate 128K output limit.

Found an error, or hit a problem this article does not cover? Send the article URL and what you saw to hi@laozhang.ai. We check it and update the article.