Gemini 4 Argon leads Google’s benchmark table on 13 of 19 rows against GPT-6 Astra, Claude Opus 5.5 and Claude Fable 5.1, but one fact outweighs any score: you can buy Astra and Opus 5.5 today, and you can’t buy Argon. Google’s new frontier model is available today only to Fairwind Program defenders, at $2/$10 per million tokens during an intro period and $4/$20 after, which is exactly what Opus 5.5 costs.
This comparison lines up the four models on price and limits, groups the benchmark rows by which model wins them, explains why the numbers aren’t apples to apples, and ends with a routing guide and a way to test all three on your own prompts in Apidog. New to Argon? Start with what is Gemini 4 Argon; for timing, see the Argon release date guide.
Price and limits side by side
Google’s table includes Claude Fable 5.1, so it gets a column too. Prices are per 1M tokens, from each vendor’s own page: Google’s launch post, OpenAI’s GPT-6 Astra model page and Anthropic’s pricing page.
| Gemini 4 Argon | GPT-6 Astra | Claude Opus 5.5 | Claude Fable 5.1 | |
|---|---|---|---|---|
| Can you call it today? | No, Fairwind only | Yes | Yes | Yes |
| API model ID | Not published | gpt-6-astra |
claude-opus-5-5 |
claude-fable-5-1 |
| Input | $2 intro, then $4 | $10 | $4 | $10 |
| Cached input | $0.10 intro, then $0.20 | $1 | $0.20 | $0.25 |
| Output | $10 intro, then $20 | $50 | $20 | $50 |
| Max output | 1M (Google’s stated limit) | 128K | 128K (300K on Batch, beta) | 128K |
| Context window | Not published | 1,050,000 (922K max input) | 1M | 1M |
| Long-prompt surcharge | Not stated | Over 272K input: 2x input and cache, 1.5x output, on the full request | None | None |
Three things stand out. Argon’s standard price matches Opus 5.5 on input, cached input and output, so its price edge over Anthropic lasts only as long as the intro period, and Google hasn’t dated that. Astra and Fable 5.1 list at 2.5x Argon’s standard input and output rates and 5x its intro rates. And the 1M output figure is Google’s: Vals AI lists a 262K max output for the Argon configuration it tested. For per-request cost math, see Gemini 4 Argon pricing.
Where each model leads in Google’s table
Before the numbers: this is Google’s table. Argon’s scores are Google-reported, some self-computed and some from leaderboards; competitor scores are mostly the vendors’ own figures or public-leaderboard results, at maximum reasoning settings where available. Harnesses differ, so read small gaps as noise.
| Who leads | Benchmark | Argon | Astra | Fable 5.1 | Opus 5.5 |
|---|---|---|---|---|---|
| Argon: knowledge work | Vals Index | 68.9% | 63.1% | 65.8% | 67.0% |
| Argon: knowledge work | AutomationBench | 51.3% | 41.4% | 31.4% | 42.5% |
| Argon: knowledge work | Harvey’s Legal Agent Benchmark | 19.6% | 5.4% | 6.7% | 3.8% |
| Argon: coding | DeepSWE v1.1 | 77.9% | 74.1% | 67.4% | 74.2% |
| Argon: long context | GraphWalks 256K to 1M (F1) | 84.2% | 71.8% | 65.0% | 66.8% |
| Argon: multimodal | LVBench | 91.7% | 87.5% | 79.7% | 83.7% |
| Argon: multimodal | Chartography | 71.6% | 71.0% | 46.2% | 66.3% |
| Astra | FrontierSWE v2 | 55.0% | 65.5% | 56.3% | 62.3% |
| Astra | Terminal-Bench Science 0.1 | 57.6% | 68.1% | 52.6% | 63.3% |
| Astra | OSWorld-2.0 (offline, partial) | 69.2% | 72.6% | not reported | not reported |
| Opus 5.5 | Terminal-bench 4.0 | 57.4% | 58.2% | 57.9% | 66.4% |
| Opus 5.5 | PostTrainBench | 45.3% | 44.3% | 40.2% | 49.3% |
| Tie | CWE-bench v1 | 68.0% | 68.0% | 58.0% | 67.0% |
Argon’s other outright wins are Vals Finance Agent v2, Vibe Code Bench, LABBench 2, RiemannBench, GraphWalks up to 128K and Agent’s Last Exam. Fable 5.1 leads no row. On CWE-bench v1, DeepMind’s cyber leaderboard shows a three-way tie at 68% between Argon, Astra and Grok 4.7, with Opus 5.5 at 67%.
On Gemini 4 Argon vs Fable 5.1, Argon scores higher on 15 of the 17 rows where both report a number; Fable edges it only on FrontierSWE v2 (56.3% vs 55.0%) and Terminal-bench 4.0 (57.9% vs 57.4%). Anthropic’s own numbers are in our Claude Fable 5.1 benchmarks post, and the full 19-row table lives in Gemini 4 Argon benchmarks.
Three caveats before you trust the gaps
Competitor numbers aren’t Google’s runs. Google’s methodology says results for non-Gemini models “are sourced from providers’ self reported numbers unless otherwise mentioned,” and several rows come from public leaderboards run by Vals AI, Proximal and Surge.
Harnesses differ. On DeepSWE v1.1, Google self-computed Argon’s 77.9% with a mini-swe-agent harness, while Astra’s score comes from the public leaderboard and the Anthropic scores come from system cards. On LVBench, Gemini sampled video at 1 frame per second, while Astra got 800 frames, Opus 5.5 600 and Fable 5.1 300, “due to API limitations.”
The same model scores differently across charts. Opus 5.5’s Terminal-bench 4.0 number differs between charts depending on harness and effort setting; Anthropic reports its 66.4% at xhigh effort. So never mix sources in one table.
What third-party evaluators say
Independent leaderboards narrow the gap. Artificial Analysis lists Argon at #8 of 223, but that list counts each reasoning setting separately. By distinct model, Argon (53) ties GPT-6 Astra and Claude Fable 5.1 and sits behind Claude Opus 5.5 (58 at max) and Claude Sonnet 5.5 (56). AA also lists Argon’s hallucination rate on AA-Omniscience at 15%, against 51% for Astra at its max setting.
On Arena, Argon ranks first in Text at 1525, marked Preliminary on 4,942 votes, with Opus 5.5 fourth. Vals AI ranks it first of 41 models on the Vals Index, the first Gemini model to top it.
Not everyone inside Google is convinced: Bloomberg reported that some employees with access say Argon underwhelms on some coding and front-end design work relative to its benchmarks, a characterization Google called inaccurate.
Which one to route to
There’s no single best frontier model in 2026. Route by task, and keep the Argon column for the day it ships:
| Task | Route today | When Argon ships |
|---|---|---|
| Legal, finance and office automation agents | Opus 5.5 (67.0% Vals Index) | Test Argon: it leads all four knowledge-work rows |
| Terminal-heavy coding agents | Opus 5.5 (66.4% Terminal-bench 4.0) | Opus 5.5 still leads |
| Agentic software engineering | Astra (65.5% FrontierSWE v2) or Opus 5.5 | Argon leads DeepSWE but trails FrontierSWE; test both |
| Computer use and GUI agents | Astra (72.6% OSWorld-2.0) | Astra leads OSWorld; Argon leads Agent’s Last Exam |
| Reasoning over 256K+ token prompts | Opus 5.5 (no surcharge) or Astra (surcharge over 272K) | Argon (84.2% GraphWalks 256K to 1M) |
| Video and chart understanding | Astra (87.5% LVBench) | Argon (91.7%), with the frame-count caveat |
| Science and ML engineering | Astra (Terminal-Bench Science), Opus 5.5 (PostTrainBench) | Argon leads LABBench 2 and RiemannBench; test it |
| Single responses over 128K tokens | Opus 5.5 on Batch (300K, beta) | Argon, up to Google’s stated 1M |
| High-volume, cost-sensitive work | Opus 5.5 ($4/$20) | Argon at $2/$10 while the intro lasts |
If price matters more than peak scores, our GPT-6 Sol vs Claude Opus 5.5 comparison covers OpenAI’s cheaper Sol line against Opus.
Compare all three on your own prompts
Vendor tables can’t tell you how each model handles your prompts. A small eval in Apidog can, and you can build it today.
- Create one project with three requests: GPT-6 Astra, Claude Opus 5.5 and a Gemini request whose model field reads
{{GEMINI_MODEL}}, set togemini-3.8-flashuntil Google publishes Argon’s ID. Our GPT-6 Astra API guide and what is Claude Opus 5.5 cover each vendor’s request format. - Store each vendor’s API key as an environment variable, and put the prompt in one shared variable so all three requests get identical input.
- Add assertions: status 200, a non-empty answer, output tokens under a ceiling, and a computed cost under your budget at each vendor’s published rates.
- Run the three as one test scenario and compare each request’s results in the test report.
The cost assertion is plain arithmetic. For a request with 20,000 input and 4,000 output tokens, Astra costs 20,000 x $10/1M + 4,000 x $50/1M = $0.20 + $0.20 = $0.40. Opus 5.5 costs $0.08 + $0.08 = $0.16. Argon would cost $0.08 at intro rates and $0.16 at standard rates. Per-token price isn’t per-task cost, though: Artificial Analysis measured Argon at about 62K output tokens per task against about 27K for Astra at max effort, so measure output tokens on your own prompts.
When Argon ships, change GEMINI_MODEL, rerun the scenario, and you have a same-prompt comparison against the two models you can buy now.
FAQ
Is Gemini 4 Argon better than GPT-6 Astra? They tie at 53 on Artificial Analysis. In Google’s table Argon scores higher on most rows, but Astra wins FrontierSWE v2, Terminal-Bench Science 0.1 and OSWorld-2.0, and Astra is the one you can call today.
Gemini 4 Argon vs Claude Opus 5.5: which is better? Google’s table favors Argon on most rows; Opus 5.5 wins Terminal-bench 4.0 and PostTrainBench and leads Artificial Analysis 58 to 53. At standard rates they cost the same.
Is Gemini 4 Argon cheaper than Claude Opus 5.5? During the intro period it’s half the price ($2/$10 vs $4/$20). After that the list prices are identical, and Google hasn’t said when the intro period ends.
Which model has the largest output limit? Argon, at a stated 1M tokens, though Vals AI lists 262K for the configuration it tested. The others cap synchronous output at 128K. Our 1M output tokens guide covers what that means for your client.
Can I use Gemini 4 Argon today? Only through Google’s Fairwind Program, for a set of vetted cyber defense partners. Paid API customers and Google AI Ultra subscribers come next, with no date.
Pick by workload, then test
Argon looks strongest on knowledge work, long context and multimodal rows; Astra on FrontierSWE, Terminal-Bench Science and OSWorld; Opus 5.5 on terminal work and ML engineering, at the price Argon will charge after its intro. Build on the two you can buy now, keep the Gemini model in a variable, and rerun your scenario the day Argon opens. To set up the three-request comparison in one project, download Apidog.
