
Qwen4-Max vs GLM-5.5: One Has a Rate Card Today, the Other Is a Version Number
- GoogleNEWGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 per 1M tokens · 90 tok/s
- AnthropicNEWAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 per 1M tokens · 96 tok/s
- orcaNEWOrca: OrcaC2 1.02026-11-0717 tok/s
- OrcaNEWOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 87 tok/s
- OrcaNEWOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 133 tok/s
- OrcaNEWOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$4.00 / $20.00 per 1M tokens · 72 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 130 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 65 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 92 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 43 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Intelligence
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 per 1M tokens · 193 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Intelligence
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 per 1M tokens · 152 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Intelligence
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
- OrcaOrca: OrcaCyber Zero 1.52026-10-10$3.00 / $7.50 per 1M tokens · 119 tok/s
- openaiOpenAI: GPT-6.1 Sol2026-09-2955Intelligence
- anthropicAnthropic: Claude Sonnet 5.52026-09-2855Intelligence
- typesafeTypeSafe: Jev 1.132026-09-24$0.04 / $0.00 per 1M tokens · 219 tok/s
At 09:46 on 16 November 2026, Alibaba's Model Studio documentation replaced the Qwen4-Max-Preview page with a page headed "qwen4-max (Latest Model)", carrying a model invocation ID of qwen4-max, a per-million-token rate card, and availability in six regions. Five days earlier the same model had been a fifteen-day rental, sold only through a Credits subscription and scheduled for retirement on 25 November. GLM-5.5, the name on the other side of this comparison, has no equivalent moment to point at: as of today Z.ai has published no model card, no API identifier, no pricing row and no weights under that name, and its own documentation index still lists GLM-5.2, GLM-5.3 and GLM-5.3-Flash as the whole line. So this is not a matchup between two flagships. It is a matchup between a flagship that acquired a price this morning and a version number that has not acquired anything.
That asymmetry is the useful thing to write about, because both names are being searched as if they were peers. What follows separates them: what Qwen4-Max actually became today, what GLM-5.5 is, and what the comparison you can genuinely run — Qwen4-Max against the GLM that ships — looks like on price, context and evidence.

What changed at 09:46 this morning
The preview is still real and still separate. qwen4-max-preview, opened at 10:00 UTC+8 on 11 November, remains a Token Plan-only model with a published retirement at 24:00 on 25 November. The new qwen4-max is a different identifier with a different commercial shape: pay-as-you-go, published rates, six regions including Singapore, Frankfurt, Virginia and Tokyo, and no expiry date on the page. Alibaba's own release list dates the model to 16 November 2026 — today — and the documentation page carries the timestamp 2026-11-16 09:46:08.

Two things did not arrive with it. There is no benchmark table anywhere on the vendor's pages, vendor-reported or otherwise, and there is no entry for the model on Artificial Analysis: the leaderboard's model index has no Qwen4 row at all, and its per-model URL returns a 404. Everything below about capability is therefore documentation, not measurement, and it is labelled that way.
The rate card, as published
These are Alibaba's own figures, in US dollars per million tokens, taken from the billing page on 16 November 2026:
• Singapore, standard — $1.50 input, $4.50 output
• Beijing, Hong Kong, Frankfurt, Virginia and Tokyo, standard — $1.413 input, $4.24 output
• Cache write, five-minute TTL — $1.875 in Singapore, $1.767 elsewhere
• Cache write, one-hour TTL — $3.00 in Singapore, $2.827 elsewhere
• Cache read — $0.075 in Singapore, $0.071 elsewhere
• Batch, US region only — $0.707 input and $2.12 output for batch file work
• Free quota — one million tokens for 90 days, Singapore only
The Singapore number is the one to quote at people, because it is the international list price and it is the highest. It is also, notably, the cheapest output rate any current flagship from the big three labs charges: Qwen3.8-Max lists at $2.00 and $6.00, Kimi K3.1 Pro at $2.00 and $10.00, Grok 4.7 at $2.00 and $6.00. Qwen4-Max undercuts all of them on both sides of the meter while claiming a larger context window than any of them.
GLM-5.5, checked the same way we checked it in October
A claim that a model does not exist decays the moment it is written, so the check is worth restating rather than asserting. Z.ai's developer documentation index, fetched today, enumerates GLM-5.2, GLM-5.3 and GLM-5.3-Flash. The GLM-5.4 and GLM-5.5 documentation paths return "Page Not Found". The vendor's model-api page names GLM-5.3, GLM-5.3-Flash, GLM-5 and GLM-5V-Turbo, and nothing between 5.3 and 6. A search for the name returns Z.ai's own pages listing 5.3 as the current flagship and no announcement of a successor.
This blog published a full audit of the name on 13 October 2026, across nine surfaces, and reached the same conclusion. Nothing in the five weeks since has moved it. The distinction that matters for a buyer is not "GLM-5.5 is obscure" — it is "GLM-5.5 is a number in a naming sequence, and the models you can actually call are one digit behind it."
The comparison that exists: Qwen4-Max against GLM-5.3
GLM-5.3 is a serious model and the honest stand-in. It launched on 14 August 2026, opened its API on 18 August, and published its 743-billion-parameter weights on 28 August under a custom Z.ai licence. It is a mixture-of-experts design activating roughly 40 billion parameters per token, with a 1M-token context window and a 128K output ceiling. Here is the contrast that a reader searching for either name actually needs:
• Input price — Qwen4-Max $1.50 per 1M in Singapore, $1.413 elsewhere, versus GLM-5.3 $1.40 list
• Output price — Qwen4-Max $4.50, versus GLM-5.3 $4.40
• Context — Qwen4-Max 1,000,000 tokens versus GLM-5.3 1,000,000 tokens, a genuine tie
• Maximum output — Qwen4-Max 131,072 tokens non-thinking, with thinking and answer sharing 262,144, versus GLM-5.3 128,000 tokens
• Modality — Qwen4-Max accepts text, image and video and returns text versus GLM-5.3, whose flagship is documented as text-only, with multimodal work pushed to GLM-5.3-Flash
• Weights — Qwen4-Max publishes none, and its documentation explicitly marks fine-tuning and model deployment as unsupported, versus GLM-5.3's open weights under a custom licence
• Independent score — Qwen4-Max has none, versus GLM-5.3 at 31.7 on the Artificial Analysis Intelligence Index v5.0.1 at maximum reasoning effort, and 23.1 at low effort
• Availability — Qwen4-Max in six Alibaba regions versus GLM-5.3 first-party plus cloud resale
Read the price rows twice, because the gap is narrower than the marketing around either model implies: on input, Qwen4-Max costs seven cents more per million in Singapore and one and a third cents more in every other region. You are not paying a premium for the newer name. You are paying roughly the same money for a materially wider input modality and a slightly higher output ceiling, and giving up the one thing GLM-5.3 has that Qwen4-Max does not — weights you can download.
What the missing score is worth
GLM-5.3's 31.7 on Index v5.0.1 is a middling number for a flagship, and it sits below Kimi K3.1 Pro's 46.5 on the same vintage. That makes the Qwen4-Max comparison unusually uncomfortable in one direction: the model with no measurement is being sold against a model whose measurement is unflattering, and there is no way to know from public data whether the newer model clears it. Alibaba's documentation claims "deep reasoning", a thinking budget defaulting to 131,072 tokens and a reasoning-effort default of high — configuration, not capability. Until an independent harness scores it, the honest position is that Qwen4-Max's advantage over GLM-5.3 is documented and its disadvantage is unquantified.
For a team making a decision this week rather than next quarter, that argues for a routing posture rather than a migration. Keep GLM-5.3 on the workloads where you already know its behaviour, and put Qwen4-Max on the ones where its documented shape is the reason you are interested: video input, a prompt that needs more than 128K of output, or a conversation long enough that a 1M window with a $0.075 cache read changes the arithmetic.
Running these through one endpoint
OrcaRouter lists GLM-5.3 as z-ai/glm-5.3 at Z.ai's own list price — $1.40 in and $4.40 out per million — passed through with no markup, alongside 214 other model identifiers on the same OpenAI-compatible base URL. Qwen4-Max is not in that catalogue today; it is available through Alibaba's own API, which is where its brand-new rate card is published and where the regional pricing above applies. When a vendor cuts a price, a pass-through catalogue reflects the cut the same day rather than on the next contract cycle, which is the practical argument for keeping the models you can route on one key and the ones you cannot on a watchlist.

The routing DSL is the piece that fits this particular comparison, because the decision between these two models is not one decision — it is a per-call one. A policy can send video-bearing requests to a multimodal model and text-only bulk work to whichever of the two is cheaper per finished task that week, and automatic failover keeps a provider-side degradation from taking the whole path down while you are still collecting evidence about the new model.
What would make this a real matchup
Two events, either of which would justify rewriting this page. The first is a GLM-5.5 model card: a document that carries parameters, context, price and an API identifier together, which is the signal this blog has used all year to distinguish a shipped model from a rumour. The second is a Qwen4-Max entry on an independent leaderboard — not because a score settles which model is better, but because right now one side of this comparison has a measurement and the other has a timestamp. Until one of those lands, the correct reading is the one the documentation supports: a flagship with a price, a name without a model, and a seven-cent difference on input that decides less than either vendor would like you to think.
Compared in this article2
Detected from this article · Benchmarks: Artificial Analysis · updated daily
