
Qwen 4 与 Gemini 4 Argon:决定胜负的 4.3%,其意义超过七个百分点
- Google新Google: Gemini 4 Carbon2026-11-10$2.00 / $10.00 每百万 tokens · 90 tok/s
- Anthropic新Anthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 每百万 tokens · 97 tok/s
- orca新Orca: OrcaC2 1.02026-11-0717 tok/s
- Orca新OrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 每百万 tokens · 87 tok/s
- Orca新OrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 每百万 tokens · 133 tok/s
- Orca新OrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$4.00 / $20.00 每百万 tokens · 72 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 每百万 tokens · 126 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 每百万 tokens · 65 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 每百万 tokens · 92 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 每百万 tokens · 44 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447智能
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 每百万 tokens · 193 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929智能
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 每百万 tokens · 147 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733智能
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740智能
- OrcaOrca: OrcaCyber Zero 1.52026-10-10$3.00 / $7.50 每百万 tokens · 123 tok/s
- openaiOpenAI: GPT-6.1 Sol2026-09-2955智能
- anthropicAnthropic: Claude Sonnet 5.52026-09-2855智能
- typesafeTypeSafe: Jev 1.132026-09-24$0.04 / $0.00 每百万 tokens · 219 tok/s
The scoreboard everyone reaches for in this matchup is the headline index: 42.9 for Qwen4-Max against 50.4 for Gemini 4 Argon, seven and a half points in Google's favour. But the number that should stop a deployment decision sits three columns to the right of the one everybody quotes. Gemini 4 Argon hallucinates on 4.3% of the questions it gets wrong. Qwen4-Max hallucinates on 28.4% of them. That is a factor of 6.6, and for anyone shipping output that a human reads, it is worth more than seven index points.
The two figures come from the same suite, Artificial Analysis v5.0.1, and neither is a vendor claim. Google has not published a hallucination rate for Argon, and Alibaba has not published one for Qwen4-Max. The independent measurement is the only one that exists.
两款机型销往同一细分市场
Both are million-token, image-capable, text-out flagships from companies that build their own silicon. Qwen4-Max went metered on 16 November 2026 in six Alibaba Cloud regions; Gemini 4 Argon has been on sale since 30 September 2026 through Google's API. The cards are close enough that price is not the differentiator people assume.
• Input, per million tokens — Gemini 4 Argon $2.00; Qwen4-Max $1.50 in Singapore and $1.413 in the other five regions. Google is 33% more expensive on the Singapore comparison and 41% more against the Global rate.
• 输出,每百万 token —— Argon $10.00;Qwen4-Max $4.50 和 $4.24。这是更大的差距:主要数字上达到 2.2 倍。
• Cached input — Argon $0.10 per million; Qwen4-Max $0.075 on a read, plus published write rates of $1.875 (five minutes) and $3.00 (one hour) that Google does not break out the same way.
每完成一个索引任务的成本——Argon 为 4.97 美元;Qwen4-Max 为 3.27 美元。尽管输出速率低 2.2 倍,Qwen4-Max 每项任务却只便宜 34%,因为两者的账单大部分都来自输入。
• 有效推理强度——Argon 按 high 档测量;Qwen4-Max 按 xhigh 档测量。两个模型都提供可调档位阶梯,且这两个数值都不能与另一模型的默认值直接比较。
• 上下文——Argon 上为 1,000,000 tokens;Qwen4-Max 在厂商规格卡上同样是 100 万,而追踪器记录为 983,616,因为该处的思考模式输入有上限。
• Hosts — Argon is served by Google; Qwen4-Max by Alibaba Cloud. Neither is open-weight.

按任务拆分并不有利于更便宜的那张卡。
拆解一下这 4.97 美元和 3.27 美元,令人意外的是哪一方在思考上花得更多。Argon 每个索引任务输出 116,368 个 token,其中 68,301 个用于推理;Qwen4-Max 输出 104,940 个,其中 68,914 个用于推理。推理量几乎相同。差别在于答案:Argon 每个任务写 48,067 个 token 的最终答案,而 Qwen4-Max 为 36,025 个。Argon 在可见之处更啰嗦,在不可见之处也同样啰嗦。
That is a cheerful anomaly in Gemini's favour, because it means the 2.2× output rate is buying 33% more answer text. The per-task premium of 52% is therefore smaller than the rate card suggests, and for workloads where output dominates — long-form generation, code the model writes itself — the gap narrows further. For workloads where input dominates, which is most RAG and most agent loops, Qwen4-Max's input rate does the work instead.
值得直说:这些是基于基准测试工具建模拟算的成本,并非实际账单。你的 token 构成会有所不同,两个模型之间的比率也会随之变化。
Qwen4-Max 真正领先的地方
在这场对决中,反对 Qwen4-Max 的理由很充分,但并非一致意见,而且有两行以不容忽视的方式指向相反方向。
• GDPval — Qwen4-Max 为 1,808.69,Argon 为 1,633.51。在经济导向的专业任务上,Qwen4-Max 领先一个在指数上高出它 7.5 个点的模型 10.7%。这是本次比较中最明显的一处单项逆转。
• FrontierCode 2.0 — 26.5% 对 25.7%。差距不到一个百分点;实际上打成平手,而且是在 Argon 的主要领先优势本应预示出差距的那套测试上。
• AutomationBench — 72.6% 对 77.5%。Argon 胜出,但优势为五个百分点,而且是在一个更接近办公自动化而非软件工程的测试套件上。
• Terminal-Bench 4.0 — 48.5% 对 57.1%。在这里,Argon 的领先是实打实的:在 shell 驱动的工作上领先八点五个百分点。
• HLE Diamond —— 44.9% 对 56.7%。在硬科学上相差十二个百分点。
• Terminal-Bench Science — 23.3% 对 52.9%。这是榜单上最大的差距,也是最清晰地拉开两个模型研究能力差距的一项。
• Omniscience 准确率——27.7 对 46.1,且本文开头提到的幻觉注意事项对两者均适用。
把这组结果放在一起看,规律是:Qwen4-Max 在专业判断类任务上不落下风,但在任何需要持续技术推理的任务上都明显落败。这比“落后七个指数点”是更窄也更有用的说法,而且它指向一种不同的部署方式。

为什么4.3%的幻觉率值得买单
用数字说明这种差异。如果审阅者能在三十秒内核查一项说法,而一条捏造的说法在进入生产环境后会造成十分钟的返工,那么每千条答案的预期审阅成本几乎完全由捏造占比而非错误占比决定。Argon 的 4.3% 意味着其错误集合中大约每二十三条答案就有一条需要返工;Qwen4-Max 的 28.4% 则意味着大约每三条半就有一条。
相比之下,对于任何设有人工审核员的工作流,52% 的每任务溢价都会消失,因为在几乎任何综合成本费率下,审核员的时间都远超 token 开销。对于任何没有人工审核员的工作流,这种溢价则会在另一个方向上消失,因为无人审核的编造答案不是一项成本,而是一起事故。
The honest caveat: hallucination rate is measured on a fixed question set and is a property of the measurement as much as the model. Google could change Argon's behaviour with a system-prompt revision tomorrow and this number would not move until the suite is re-run. Treat it as a strong prior, not a guarantee.
路由问题,单方面作答
Gemini 4 Argon is on our catalogue as google/gemini-4-argon at Google's list price, $2.00 in and $10.00 out per million, with 0% markup on top — the same rates Google publishes, passed through rather than repriced. Qwen4-Max is not on our catalogue: a qwen4 identifier returns not-found, its route is Alibaba's Model Studio API and several third-party platforms, and the closest Qwen model we serve is qwen/qwen3.8-max at $2.00 and $6.00.
对大多数团队来说,这种不对称本身就是部署问题的答案。在幻觉这一项上胜出的模型,是你今天就能通过已有端点调用的那个;而在 GDPval 上胜出的模型,想试用就得再签一份供应商合同。如果你想在正式投入前用自己的数据验证 GDPval 的说法,路由 DSL 让你可以用一次调用把固定的评估集同时送入两条路径,并并排比较输出——这比两个模型各自的单任务价格所暗示的实验成本都要低。

这七点没有告诉你的事
有两件事会改变这一对比,而这两件事都并不遥远。在一套为拒答而调优系统提示词的 Qwen4-Max 部署上进行独立幻觉测量,将检验 28.4% 究竟是模型本身的属性,还是查询方式造成的——尚无供应商发布过此类测量。而在更大样本上重新运行 GDPval 将告诉我们,Qwen4-Max 在专业任务上 10.7% 的领先是真实的还是噪声,因为该基准的置信区间很宽,且两个模型差距足够小,排序可能翻转。
Until either lands, the split is clean and it is not the one the index suggests. Sustained technical reasoning, science, shell work, or anything a customer reads unreviewed: Gemini 4 Argon, and the 52% premium is the least interesting number in the deal. Grounded professional judgement at volume, with a review step you were going to run anyway: Qwen4-Max is 10.7% better on the suite that measures exactly that, and it is 34% cheaper per task while you find out.
