A generated title card reading 'Qwen4-Max vs Gemini 4 Pro' with the subtitle 'The Pro slot is empty — Argon and Hydrogen are what it points at' and three figure cards reading '43 vs 50 vs 54 on the index', '$4.50 vs $10.00 vs $7.50 output' and 'no gemini-4-pro exists'. A footer reads 'Scores per Artificial Analysis Intelligence Index v5.0.1; prices per each vendor's rate card.' Flat vector styling on a white background with blue-and-cyan gradient accents, the OrcaRouter mark composited bottom right.
Guides & Insights

Qwen4-Max vs Gemini 4 Pro: Google's Pro Slot Is Empty, and the Filled Ones Cost Less Than You Think

Author

jinhao.song

Date Published

Latest models · 20View all models →
Benchmarks: Artificial Analysis · updated daily
Back to all posts

There is no model called Gemini 4 Pro. Google's own model list and pricing page carry three Gemini 4 identifiers — gemini-4-carbon, gemini-4-argon and gemini-4-hydrogen — and no Pro string anywhere; the Pro tier's last occupant is a 3.1 preview that Google has left in place while it shipped the 4 line under new names. That matters for this comparison because Qwen4-Max, priced at $1.50 per million input tokens and $4.50 per million output since its 16 November 2026 general availability, is not competing against an empty label. It is competing against Gemini 4 Argon, the model the Pro name actually points at, and against Gemini 4 Hydrogen, the one that arrives at the same input price as Qwen4-Max and beats it by eleven points on the Artificial Analysis Intelligence Index. So the honest version of this matchup has two answers, and the cheaper one is not the one the query asked for.

Why the Pro name brings people here, and what is actually behind it

Google's Gemini naming is positional. Flash is throughput, Pro is general capability, Ultra is the top when one exists. Through the 3.x generation the Flash slot moved four times — 3.5 through 3.8 — while the Pro slot never received a numbered successor beyond 3.1. When Google then announced a Gemini 4 line in September and October 2026 under element names, the Pro label was left with nothing standing behind it, and search demand routed the expectation into a string that resolves to no rate card, no model card and no endpoint.

• The model list — three Gemini 4 entries, all with 1M-token input and a 64K-to-1M output ceiling. No gemini-4-pro.

• The pricing page — rows for Carbon, Argon and Hydrogen. No Pro row, and no Pro model card exists at any point in 2026.

• The nearest thing to a Pro — Gemini 4 Argon, announced 30 September 2026, generally available on the paid API from 13 October 2026. It is the general-capability tier under a new name, and it is what every "Gemini 4 Pro benchmark" figure circulating online was actually measured on.

• The one to watch — Gemini 4 Hydrogen, shipped 12 November 2026 at $1.50 input and $7.50 output with no promotional end date, which makes it the cheapest way to buy Gemini 4 capability and, on input price, an exact match for Qwen4-Max.

• Carbon — the highest-scoring of the three at 60 on the index, announced 28 October 2026, and distributed through restricted-access programs rather than open sign-up, so it is not the model most readers can simply call today.

The three-column truth: Qwen4-Max, Argon, Hydrogen

All index figures below are Artificial Analysis Intelligence Index v5.0.1 readings taken on 16 November 2026 — the day Qwen4-Max went GA — and both vendors' prices are list rates before any negotiated discount.

• Intelligence Index — Qwen4-Max 43, ranked 38th of 81. Gemini 4 Argon 50, ranked 22nd. Gemini 4 Hydrogen 54, ranked 16th. Carbon, for completeness, 60 and fourth.

• Input price — Qwen4-Max $1.50 in Singapore and $1.413 elsewhere. Argon $2.00. Hydrogen $1.50. On input there is no gap at all between Qwen4-Max and Hydrogen.

• Output price — Qwen4-Max $4.50. Argon $10.00. Hydrogen $7.50. This is the row where Qwen4-Max wins against every live Google model.

• Price clock — Argon's $2.00 / $10.00 is introductory and doubles to $4.00 / $20.00 on 1 December 2026, with caching rates doubling too. Hydrogen's card carries no cutover date. Qwen4-Max's card carries none either.

• Cost per Intelligence Index task — Qwen4-Max $3.27. Argon $4.97. Hydrogen is not published with a per-task figure in the same block, which is a gap rather than a win.

• Speed, where it exists — Argon 66 output tokens per second, 45.59 seconds to first answer token, 29.3 minutes per index task. Qwen4-Max has no performance reading at all yet: the tracker shows N/A for time per task and blanks for speed. That empty column is the largest single unknown in this matchup.

• Context and output — all three at 1M tokens of input. Argon's ceiling is documented as up to 1M output on long-decode work; Qwen4-Max caps non-thinking output at 131,072 and puts thinking plus answer inside 262,144 shared tokens.

• Input modality — Qwen4-Max takes text, images and video. The Gemini 4 pages list text and image for Argon and Hydrogen. Video is a real, if narrow, Qwen advantage.

A generated two-column scoreboard titled 'Qwen4-Max vs Gemini 4 Hydrogen — the scoreboard'. The Qwen4-Max column reads Input price $1.50 per million, Output price $4.50 per million, AA Index 43 rank 38 of 81, Price per index task $3.27, Context 1,000,000 tokens, Announced 16 November 2026. The Gemini 4 Hydrogen column reads Input price $1.50 per million, Output price $7.50 per million, AA Index 54 rank 16 of 81, Price per index task not published, Context 1,000,000 tokens, Announced 12 November 2026. A footer reads 'Index figures per Artificial Analysis v5.0.1; no model named Gemini 4 Pro exists.'

What the date of writing does to the price comparison

Two weeks separate this page from a completely different answer, and the arithmetic is worth doing before it happens rather than after.

Today, Qwen4-Max at $1.50 / $4.50 against Argon at $2.00 / $10.00 is a 25% input and 55% output saving, and Argon is seven index points ahead. From 1 December 2026, Argon bills $4.00 / $20.00 unless Google extends the promotion, and the same comparison becomes 62.5% cheaper on input and 77.5% cheaper on output for Qwen4-Max — for the same seven-point deficit. A team that evaluates in late November and signs in December is evaluating a different deal from the one it gets.

Hydrogen is the awkward one, because it does not move on 1 December and it is already the strongest model in the comparison. At $1.50 input it matches Qwen4-Max to the cent, costs $3.00 more per million output tokens, and scores eleven points higher on the same index revision. For any workload where output is a small share of the bill — retrieval, classification, extraction, long-document QA — Hydrogen is simply the better purchase, and no amount of Qwen efficiency arithmetic fixes an eleven-point capability gap on a 0% input premium. Where output dominates, which is true of most agentic and generative work, Qwen4-Max's $4.50 against $7.50 restores a 40% saving on the expensive side of the meter and turns the question back into one about quality per dollar.

Where Qwen4-Max's advantage is real, and not about price

Three things in the launch material are not marketing. The first is the budget structure: 1,000,000 tokens of context with a documented 991,808-token input ceiling in non-thinking mode and 983,616 in thinking mode, on a model whose vendor explicitly publishes both numbers rather than one headline. The second is protocol breadth — OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and DashScope, all four, with structured output, parallel tool calling, prefix completion and both explicit and implicit caching, which means an existing integration can move its request body without a rewrite. The third is batch support in Beijing, where Batch File bills $0.707 / $2.12 per million; it is one region out of six, and it is priced lower than anything Google publishes for asynchronous work.

None of that changes the index ordering. What it does mean is that the choice between these models is closer to a platform decision than the score gap implies, and that a team already standardized on Gemini's wire format has less reason to move than the price table alone would suggest — while a team already speaking OpenAI or Anthropic formats could call Qwen4-Max without touching its client code.

A headless-browser screenshot of the Alibaba Cloud Model Studio documentation page for qwen4-max (Latest Model), captured 16 November 2026 in English, showing the model invocation ID 'qwen4-max', a breadcrumb under Get Started, a description of Qwen4-Max as the next-generation flagship multimodal model covering text generation, image and video understanding and deep reasoning, and a note that it is compatible with the OpenAI and Anthropic APIs; rate-limit, thinking-effort, region and model-pricing entries are visible in the page's left navigation.

Getting Argon and Hydrogen through one key

Gemini 4 Argon is routed on OrcaRouter as google/gemini-4-argon at Google's own list price with 0% markup — $2.00 in and $10.00 out per million tokens today, with the 1 December repricing passed through the day Google publishes it, not after a reseller cycle. Gemini 4 Hydrogen, shipped 12 November 2026, is not in that catalogue as of 16 November — it is at Google's own API and several third-party platforms, and the same is true of Qwen4-Max through Alibaba's Model Studio. The practical value of the pass-through applies to whatever is routed here: when Argon's introductory rate expires on 30 November, an OrcaRouter bill reflects the new rate on 1 December, and the comparison you ran in November stops describing your invoice.

Qwen4-Max is likewise not routed here as of 16 November 2026, so of the three models in this piece only the Argon arm needs no second contract. That asymmetry is worth stating plainly, because the cheapest honest test of the Qwen4-Max case — route a shadow slice at it, price per completed task rather than per token, hold the incumbent on the volume — costs one integration on one side and none on the other.

A headless-browser screenshot of the OrcaRouter model page for google/gemini-4-argon, captured 16 November 2026 in English, showing the Gemini 4 Argon heading and identifier, the Google attribution, OpenAI- and Gemini-compatible sample code pointed at api.orcarouter.ai/v1, the pricing panel reading $2.00 input and $10.00 output per million tokens, a 10.60 s measured latency, 97.2M tokens served and a zero-data-retention note.

The number to take away

If you arrived looking for Gemini 4 Pro, the model you want is Gemini 4 Argon, and it is seven points ahead of Qwen4-Max for about twice the output price until December and four times the output price after. If what you actually wanted was the strongest Gemini 4 against a much cheaper Chinese flagship, the model you want is Gemini 4 Hydrogen, and the answer is less flattering to the cheaper side: eleven points ahead at an identical input price, with a 40% output premium that only pays off if your workload writes a lot.

The one figure that could still move all of it is the blank cell. Artificial Analysis has scored Qwen4-Max at 43 and measured nothing else about how it runs — no output speed, no time to first token, no minutes per task. Argon's 29.3 minutes per task and Hydrogen's speed are known quantities, and a model that is 40% cheaper per output token but materially slower is not 40% cheaper on anything with a human waiting. Treat the Qwen4-Max price advantage as real and the Qwen4-Max performance claim as not yet made, and revisit both when the tracker fills that column in.

Compared in this article1

Detected from this article · Benchmarks: Artificial Analysis · updated daily