APIMODELSAPIMODELS

Qwen 4 Max vs GPT-6 Sol

TL;DR Qwen 4 Max and GPT-6 Sol are level on overall intelligence, and GPT-6 Sol is faster and uses far fewer tokens to get there. Artificial Analysis scores Qwen 4 Max at 43 and GPT-6 Sol at 44; GPT-6.1 Sol, the upgrade at the same list price, scores 55. By Alibaba’s own chart Qwen leads GPT-6.1 Sol on nine of sixteen benchmarks, mostly agent, automation and legal work, and trails on terminal coding, charts and video. In our seven same-prompt tests all three scored 93 out of 93; GPT-6 Sol finished in 189 seconds and Qwen 4 Max in 415, and the bill was $0.09 against $0.21. List prices are $1.50 / $4.50 per million tokens for Qwen 4 Max and $2 / $10 for GPT-6 Sol. All three run on one apimodels.app key.

Qwen 4 Max is Alibaba’s new flagship, open for pay-as-you-go API use since 16 November 2026 under the model ID qwen4-max. GPT-6 Sol is the workhorse tier of OpenAI’s GPT-6 family, released in September 2026, and it sits in the same price band, which is why the two get compared.

One thing to know before reading the tables: OpenAI released GPT-6.1 Sol a few days after GPT-6 Sol at the same list price, and most third-party tests, including Alibaba’s own launch chart, compare Qwen 4 Max with the 6.1 version. We show both and label which is which. We also ran all three ourselves on the same seven tasks.

We run apimodels.app, which sells all three through one API key, so we have no reason to favour any of them. The benchmark scores below come from Artificial Analysis, Vals AI, Alibaba and OrcaRouter, with links at the end. The seven same-prompt tests are ours, run on 17 November 2026 through our own API with each model’s default thinking setting.

Qwen 4 Max

Stronger on agent and legal tests, one price at any length

  • ·Artificial Analysis Intelligence Index 43
  • ·Leads GPT-6.1 Sol on APEX-Agents 72.1 vs 60.0 and HLAB legal 21.3 vs 6.3 (Alibaba’s chart)
  • ·$1.50 / $4.50 per 1M tokens with no long-prompt surcharge
  • ·Text, image and video input
Qwen 4 Max price and API →

GPT-6 Sol

Same level, about twice as fast, fewer tokens

  • ·Artificial Analysis Intelligence Index 44; GPT-6.1 Sol 55
  • ·GPT-6.1 Sol leads on Terminal-Bench 4.0 56.1 vs 48.5 and ProgramBench 52.5 vs 31.5
  • ·$1.00 / $5.00 per 1M tokens on apimodels.app (list $2 / $10)
  • ·Text and image input; reasoning effort from none to max
GPT-6 Sol price and API →

Pick GPT-6 Sol (or 6.1 Sol) when a person is waiting, for terminal and repository coding, and for charts and screenshots. Pick Qwen 4 Max for agent and automation pipelines, legal and document work, video input, and prompts longer than 272K tokens.

Specs and prices side by side

The two have near-identical context windows and output caps. The pricing structure is where they differ. GPT-6 Sol doubles its input price and raises its output price by half for any request longer than 272,000 tokens; Qwen 4 Max charges one rate across the whole window. For very long prompts that makes Qwen’s input about 2.7 times lower and its output about 3.3 times lower at list prices.

Thinking works differently too. Qwen 4 Max thinks by default and lets you switch it off per request. GPT-6 Sol has a reasoning effort setting with six levels from none to max, and medium is the default.

Qwen 4 MaxGPT-6 SolGPT-6.1 Sol
Maker, releaseAlibaba, 16 Nov 2026OpenAI, Sep 2026OpenAI, 29 Sep 2026
List price per 1M tokens (input / output)$1.50 / $4.50$2 / $10$2 / $10
Cached input, list$0.075$0.20$0.10
Requests over 272K tokens, listNo change$4 / $15$4 / $15
Price on apimodels.app$1.50 / $4.50$1.00 / $5.00$1.00 / $5.00
Context window1M1.05M1.05M
Max output131K (262K with thinking)128K128K
Input typesText, image, videoText, imageText, image
Thinking controlOn by default, can be switched offEffort none to max, default mediumEffort none to max

Independent benchmarks

On Artificial Analysis’s Intelligence Index, Qwen 4 Max scores 43 and GPT-6 Sol 44: a tie. GPT-6.1 Sol is a clear step up at 55 with maximum effort and 46 at medium. On the Vals Index the picture is the same from another lab: Qwen 4 Max at 63.20% and GPT-6.1 Sol at 63.94%.

Where they separate is in what it takes to get those scores. Qwen 4 Max wrote 127 million output tokens to finish the Artificial Analysis index; GPT-6 Sol wrote 65 million and GPT-6.1 Sol 64 million. Half the tokens means the Sol models finish sooner and, despite a higher price per token, cost less per task: $2.47 for GPT-6 Sol and $1.27 for GPT-6.1 Sol against $3.27 for Qwen at list prices.

MeasureRun byQwen 4 MaxGPT-6 SolGPT-6.1 Sol
Intelligence Index v5.0.1Artificial Analysis434455 (max); 46 (medium)
Output tokens to run the indexArtificial Analysis127M65.1M63.7M
Cost per index task, list pricesArtificial Analysis$3.27$2.47$1.27
Output speedArtificial Analysisnot measured yet125 tokens/s49 tokens/s
Vals Index v2.2Vals AI63.20%not listed63.94%
Cost per Vals testVals AI$9.33not listed$3.61

What Alibaba’s own chart shows (against GPT-6.1 Sol)

Alibaba’s launch chart puts Qwen 4 Max next to GPT-6.1 Sol on sixteen benchmarks. By Alibaba’s numbers Qwen leads on nine and GPT-6.1 Sol on seven. These are vendor-reported, so treat them as Alibaba’s claim.

Qwen’s wins cluster around agents that work through business tasks: APEX-Agents 72.1 against 60.0, AutomationBench 72.5 against 64.9, the Briefcase office-work ranking 1686 against 1555, and legal work on HLAB at 21.3 against 6.3. GPT-6.1 Sol’s wins are terminal and repository coding (Terminal-Bench 4.0 56.1 against 48.5, ProgramBench 52.5 against 31.5) and anything visual: charts 67.8 against 54.2 and video understanding 73.8 against 68.1, even though Sol cannot take a video file directly through the API.

Benchmark (reported by Alibaba)Qwen 4 MaxGPT-6.1 Sol
DeepSWE v1.175.775.2
CodeArena (Elo)17501742
Agents’ Last Exam v1.146.742.6
AA-Briefcase v1.1 (Elo)16861555
APEX-Agents v1.172.160.0
AutomationBench-AA72.564.9
HLAB-Vals-1.1 (legal)21.36.3
JobBench68.566.2
BioMysteryBench-1.185.084.4
Terminal-Bench 4.048.556.1
ProgramBench31.552.5
AgentArena net gain7.511.8
Humanity’s Last Exam51.754.0
HealthBench Professional60.065.1
Chartography54.267.8
Video-Index68.173.8

Our seven same-prompt tests: same answers, half the time

We wrote seven tasks modelled on what developers most often send through our API: writing a function, fixing a buggy class, a support agent that calls tools and has to follow a refund policy, questions over a 1,500-line production log, turning a Chinese group chat into JSON, three counting puzzles, and Chinese product copy with hard constraints. Each has an automatic check, 93 checks in total, verified against reference answers before we ran anything.

Qwen 4 Max, GPT-6 Sol and GPT-6.1 Sol each got all 93 checks right. All three passed the 50 hidden cases on the function, found the five planted bugs, refunded only the defective keyboard for $129 and turned down the 46-day-old mouse return, and extracted six orders from the chat with the right dates for “before Friday” and “next Wednesday”.

The gap is in time and tokens. GPT-6 Sol finished the set in 189 seconds and GPT-6.1 Sol in 197; Qwen 4 Max took 415. The long log shows it most clearly: GPT-6 Sol answered all five questions correctly in 6 seconds, Qwen needed 63. On the three puzzles the Sol models reasoned for about a minute and Qwen for more than three.

Qwen wrote about five times as many output tokens, 22,100 against 4,350 for GPT-6 Sol, almost all of it thinking. At apimodels.app prices the seven tasks cost $0.21 on Qwen 4 Max and $0.09 on either Sol model.

On the one task without a right answer, the Chinese product copy, all three met the five constraints. Qwen’s copy was the most specific and also the most invented: it added a 250-gram weight, bone-conduction pickup and a fast-charge claim we never supplied. GPT-6 Sol stayed closest to what the prompt gave it.

Task (checks)Qwen 4 MaxGPT-6 SolGPT-6.1 Sol
Write a timeline-merge function (50 hidden cases)50/50 · 51 s · $0.01250/50 · 40 s · $0.00550/50 · 39 s · $0.004
Fix a TTL cache class with 5 planted bugs (10)10/10 · 27 s · $0.00910/10 · 41 s · $0.00410/10 · 16 s · $0.004
Refund agent with tools and a policy (7)7/7 · 16 s · $0.0067/7 · 12 s · $0.0127/7 · 14 s · $0.015
Five questions over a 1,500-line log (5)5/5 · 63 s · $0.1205/5 · 6 s · $0.0545/5 · 28 s · $0.054
Chinese group chat to JSON (13)13/13 · 19 s · $0.00413/13 · 9 s · $0.00313/13 · 12 s · $0.003
Three counting puzzles (3)3/3 · 196 s · $0.0493/3 · 57 s · $0.0053/3 · 63 s · $0.004
Chinese product copy, five hard constraints (5)5/5 · 43 s · $0.0105/5 · 23 s · $0.0075/5 · 25 s · $0.004
All seven93/93 · 415 s · $0.2193/93 · 189 s · $0.0993/93 · 197 s · $0.09

What a task costs

This is the comparison where price per token misleads most. Qwen 4 Max is priced about 2.2 times lower per token than GPT-6 Sol at list ($1.50 / $4.50 against $2 / $10). But it writes roughly twice the output tokens for the same work, and output is the expensive side. OrcaRouter’s analysis put a typical task about 32% higher on Qwen 4 Max than on GPT-6 Sol, and Artificial Analysis’s own figures agree: $3.27 against $2.47 per task.

On apimodels.app GPT-6 Sol and GPT-6.1 Sol are $1.00 / $5.00, half of OpenAI’s list, which brings an Artificial Analysis task to about $1.24 on GPT-6 Sol and about $0.64 on GPT-6.1 Sol. Qwen 4 Max is at Alibaba’s list price, about $3.27. In our own lighter set the ratio was similar: $0.21 against $0.09.

There are two cases where Qwen 4 Max comes out ahead on cost. The first is very long input: above 272,000 tokens GPT-6 Sol’s list price doubles to $4 for input and rises to $15 for output, and Qwen’s stays put. The second is short-answer work with thinking switched off, where Qwen’s lower output price applies without the thinking overhead.

Speed

GPT-6 Sol is the faster model by a wide margin. Artificial Analysis measured 125 output tokens per second for GPT-6 Sol and 49 for GPT-6.1 Sol; it has not published a figure for Qwen 4 Max yet, OpenRouter’s live data shows about 30, and reviewers of the preview measured 33 to 53.

On our platform over the last 30 days GPT-6 Sol returned its first token after about 5 seconds at the median (production calls, sample size not disclosed). Qwen 4 Max went on sale here on 16 November, so there is no comparable production figure yet; in our tests, leaving out the puzzles where both think longest, its first visible word took 15 to 62 seconds, against 3 to 35 for GPT-6 Sol.

Which to use

Chat, support and anything where a person waits: GPT-6 Sol. It was about twice as fast across our tests and ten times as fast on the long log.

Terminal and repository-scale coding: GPT-6.1 Sol, which leads Qwen on Terminal-Bench 4.0 and ProgramBench by a wide margin. For single functions and bug fixes all three were perfect in our tests.

Agents that carry out business processes, automation, legal and contract work: Qwen 4 Max. APEX-Agents, AutomationBench, Briefcase and HLAB are where it leads, and the legal margin is its largest.

Charts, screenshots and other image reading: GPT-6.1 Sol, 67.8 against 54.2 on Chartography.

Video files as input: Qwen 4 Max; the Sol models take text and images only.

Prompts longer than 272K tokens: Qwen 4 Max, because its price does not change with length.

If you are choosing between the two Sol versions, GPT-6.1 Sol scores higher on most tests at the same price; GPT-6 Sol is the quicker of the two. On apimodels.app all three are one key and a change of the model string.

Caveats

Qwen 4 Max has been generally available for one day at the time of writing; Artificial Analysis has run only one thinking setting for it and has not measured its speed. The vendor chart compares it with GPT-6.1 Sol, not GPT-6 Sol, and no third party has yet published a same-task test of Qwen 4 Max against either Sol version on Chinese-language work or on hallucination.

Our seven tasks are everyday-difficulty work, run once per model at default settings. They show all three are dependable at that level and that the Sol models get there faster; they do not measure the hard end.

Sources

Specs and list prices: Alibaba’s QwenCloud model page and OpenAI’s pricing and model pages, read on 17 November 2026. Independent scores, token counts, task costs and speeds: Artificial Analysis and Vals AI, read the same day. Alibaba’s sixteen-benchmark chart as reproduced by DataCamp. Cost-per-task analysis: OrcaRouter. The seven same-prompt tests are our own, run on 17 November 2026.

Quick start

cURL

curl https://api.apimodels.app/v1/chat/completions \
  -H "Authorization: Bearer $APIMODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen4-max",
    "messages": [{"role": "user", "content": "Summarise the open issues in this log: ..."}]
  }'
# same key, same endpoint: change "model" to "gpt-6-sol" or "gpt-6.1-sol"

FAQ

Is Qwen 4 Max better than GPT-6 Sol?

They are about level overall. Artificial Analysis scores Qwen 4 Max at 43 and GPT-6 Sol at 44, and in our seven same-prompt tests both scored 93 out of 93. Qwen 4 Max is stronger on agent, automation and legal benchmarks; GPT-6 Sol is about twice as fast and uses roughly half the output tokens. GPT-6.1 Sol, the upgrade at the same list price, scores 55.

Which costs less, Qwen 4 Max or GPT-6 Sol?

Per token Qwen 4 Max lists lower, $1.50 / $4.50 against $2 / $10 per million tokens. Per task GPT-6 Sol usually costs less because it writes about half the output tokens: Artificial Analysis measured $2.47 per task on GPT-6 Sol and $3.27 on Qwen 4 Max at list prices. Qwen comes out ahead on prompts longer than 272,000 tokens, where GPT-6 Sol’s list price rises to $4 / $15.

Which is faster, Qwen 4 Max or GPT-6 Sol?

GPT-6 Sol is faster. Artificial Analysis measured 125 output tokens per second for GPT-6 Sol, while public figures for Qwen 4 Max range from about 30 to 53. Across our seven tests GPT-6 Sol took 189 seconds and Qwen 4 Max 415.

Should I compare Qwen 4 Max with GPT-6 Sol or GPT-6.1 Sol?

Compare it with GPT-6.1 Sol if you want OpenAI’s current model at this price: it has the same $2 / $10 list price as GPT-6 Sol, a lower cached-input price, and a higher Artificial Analysis score, 55 against 44. GPT-6 Sol remains available and is the faster of the two.

Can Qwen 4 Max take video input?

Yes. Alibaba lists text, image and video input for Qwen 4 Max. GPT-6 Sol and GPT-6.1 Sol accept text and images only.

Can I use Qwen 4 Max and GPT-6 Sol with one API key?

Yes. Both run on one apimodels.app key. Call POST https://api.apimodels.app/v1/chat/completions with model qwen4-max, gpt-6-sol or gpt-6.1-sol.

Related

Мы на
APIMODELS on Product HuntAPIMODELS featured on ToolifyAPIMODELS featured on There's An AI For That