Qwen4-Max vs Hy4 Flash: Which Cheap Model Is Worth It
Qwen4-Max lists at $1.50 per 1M input tokens, Hy4 Flash at $0.07. The rate cards, the cost-per-task numbers, and which claims come from the vendors.

Two cheap models landed within a day of each other: Alibaba's Qwen4-Max at $1.50 per million input tokens, Tencent's Hy4 Flash at $0.07. Both ship a 1M-token context. Neither rate card tells you which one costs less to finish your actual work.
What shipped
Vercel's changelog, dated 16 Nov 2026, says Hy4 Flash from Tencent "is now available on AI Gateway with an introductory 50% discount on input, cached input, and output tokens through November 29, 2026." The model name on the Gateway is tencent/hy4-flash.
Vercel describes it as the new lower-cost member of the Hy4 family, keeping "the full Hy4 feature set at roughly a tenth of the price per token": reasoning with preserved thinking across turns, tool calling, structured outputs, prompt caching and image input, with the same 1M-token context window. Reasoning can be switched off for requests that do not need it — Vercel points that at high-volume coding, agent and document workloads.
Qwen4-Max arrived the same day. According to a Dev.to roundup by Muhammad Ali Mughal, who also publishes a daily AI news video on the channel Daily AI News By Ali and says he checked every number against the linked primary sources himself, Alibaba's Qwen team announced the model on Nov 16. The roundup cites the official blog describing a mixture-of-experts model with 2.4T backbone parameters and 71B active per token, and the QwenCloud model page listing text, image and video input, a 1M-token context, and pricing of $1.50 per 1M input tokens, $4.50 per 1M output and $0.075 per 1M for implicit cache.
Per the same roundup, Tencent's own figures for Hy4 Flash are 224B total parameters, 12B active, a 1M context, and pricing of $0.070 per 1M input, $0.281 per 1M output and $0.003 cached.
The rate cards side by side
| Qwen4-Max | Hy4 Flash | |
|---|---|---|
| Input / 1M tokens | $1.50 | $0.070 |
| Output / 1M tokens | $4.50 | $0.281 |
| Cache | $0.075 (implicit) | $0.003 |
| Context | 1M tokens | 1M tokens |
| Parameters | 2.4T backbone / 71B active | 224B total / 12B active |
| Availability | API-only; OpenRouter and Vercel AI Gateway | Vercel AI Gateway; OpenRouter "coming soon" per Tencent |
All figures above come from the roundup's reading of the vendors' own pages, except the Gateway availability and discount, which are Vercel's. Note the two discounts are different offers in different places: Vercel's 50% cut runs on AI Gateway through November 29, 2026, while the roundup says Hy4 Flash is 50% off on Tencent Cloud TokenHub through Nov 30.
Also worth filing before you plan around it: the roundup notes the Max model is API-only, and that Qwen says open weights are coming only for the smaller Qwen4 models.
Per-token price is half the bill
The more useful framing in the roundup is cost per finished task, not cost per token. A model that burns fewer output tokens to complete a job can be cheaper even on a similar rate card.
Alibaba's own numbers make that case. Per the roundup, Alibaba's internal Work360 benchmark puts Qwen4-Max's average cost per task at $1.67, down from $4.43 for Qwen3.8-Max. That is a vendor benchmark on a vendor workload — treat it as a direction, not a measurement of your workload.
The independent reading is more restrained. The roundup cites Artificial Analysis scoring Qwen4-Max at 43 on its Intelligence Index — 9 points above Qwen3.8 Max, but 3 behind Kimi K3.1 Pro — at $3.27 per index task. The same source measured roughly 105k output tokens per task versus about 171k for Qwen3.8 Max, which is the mechanism behind the cost-per-task drop. On the leaderboard side, the roundup says Arena has it at #6 in Code Arena WebDev, preliminary.
So the cheaper-per-task story survives contact with a third party, but the quality jump is smaller than the price drop implies, and it does not put the model at the front of the field.
The parity claim is Tencent's own
Hy4 Flash's headline is a comparison, and it is worth reading carefully before you route traffic at it. The roundup reports that Tencent claims the model matches Claude Haiku 5.5 (high) on daily work in human blind tests, based on 85 internal users and 259 tasks — and the roundup itself flags this: "That is Tencent's own evaluation, so treat the parity claim as a vendor claim."
Eighty-five internal users is a small, non-independent panel. Nothing in the material establishes how the tasks were chosen or who graded them. Until an outside benchmark lands, the defensible version of that claim is that Tencent believes the two are close on everyday work, not that they are.
Tencent's own routing advice is the more actionable part. Per the roundup, Tencent recommends Hy4 Flash for everyday work — search, documents, slides, simple frontend demos — and the bigger Hy4 for long-horizon software engineering. A vendor telling you where its cheap model stops is a useful signal.
Is it worth switching?
The honest answer is that neither price difference settles anything by itself, and the material here does not contain a head-to-head benchmark of these two models against each other. What it does support is a routing decision rather than a replacement decision.
The split the roundup recommends: send summarization, extraction and drafting to a cheap model, keep a strong model for hard coding tasks, and log quality per route. That is standard practice, but the gap between $0.07 and $4.50 per million is wide enough that getting the split right is worth an afternoon.
Two practical notes from the primary sources. The roundup says Qwen's API is OpenAI-compatible, so an A/B test is a small config change. And on Vercel's side, using Hy4 Flash inside Claude Code, Codex or Cursor means installing the latest Vercel CLI, running npx vercel ai-gateway setup, and selecting tencent/hy4-flash in the agent — Vercel also states the Gateway reflects provider pricing with no markup and charges no platform fee on inference, including on Bring Your Own Key requests.
If you benchmark either model, measure cost per finished task on your own prompts. Both of today's favourable numbers — Alibaba's $1.67 and Tencent's blind-test parity — are vendor-produced, and the one independent measurement in the material is more cautious than both.
What to watch
- Independent benchmarks for Hy4 Flash, beyond Tencent's own blind test.
- OpenRouter availability for Hy4 Flash, which Tencent says is coming.
- Open weights for the smaller Qwen4 models — the roundup says the Max tier is not getting them.
- Whether the introductory discounts hold past the November 29 and November 30 end dates on their respective platforms.
More from DangMua