Claude Sonnet 5.5 vs GLM-5.5: There Is No GLM-5.5 to Compare Against

Search for this matchup and you will find pages that compare two models. One of them is real. Claude Sonnet 5.5 has a model card, an API identifier, a price row, a published retirement floor and an entry on the independent leaderboards; it shipped on September 28, 2026 and is the second model in the vendor's 5.5 generation. GLM-5.5 has none of those things. As of September 29, 2026 there is no GLM-5.5 checkpoint on Z.ai's Hugging Face organisation, no page for it on Z.ai's own model list, no row for it in the vendor's documentation, no price, and no Artificial Analysis entry. The name is a forecast that hardened into a noun.

That makes the honest version of this article shorter than the title promises and more useful than a spec sheet would be: what the name refers to, what the closest real model is, what the two would-be opponents actually score, and the specific evidence that would have to appear before "GLM-5.5" becomes something you can call. The comparison itself is not unavailable — it just is not the one the search box asked for.

What is actually there, checked today

Five places would carry a released Z.ai flagship, and all five were checked on September 29, 2026:

• Z.ai's own model list carries exactly three cards: GLM-5.3, badged as the company's latest flagship; GLM-5.3-FlashX, badged new; and GLM-5.2. No 5.4, no 5.5

• The vendor documentation's recommended-models list starts at GLM-5.3, then GLM-5.3-Flash, GLM-5.3-FlashX and GLM-5.2. No GLM-5.5 page exists at any documentation path

• The Z.ai organisation on Hugging Face lists its newest repositories as the GLM-5.3 series, created August 25, 2026. A search of the organisation's repositories returns no GLM-5.5 of any size, quantisation or variant

• Artificial Analysis has no GLM-5.5 model page to open. Its GLM coverage stops at the 5.3 generation

• OrcaRouter's catalogue has no GLM-5.5 entry of any kind, so there is nothing for us to route or to quote

What does exist is GLM-5.3: announced August 14, 2026, in market on the API around August 18, with weights published on August 25. It is text-in, text-out, runs a 1,000,000-token context window with a 128,000-token output ceiling, carries 753 billion total parameters with roughly 40 billion active, and costs $1.40 per million input tokens and $4.40 per million output tokens on Z.ai's list. Its thinking is always on — disabling it is no longer accepted — and depth is controlled through reasoning_effort at low, high or max, defaulting to max.

Where the name keeps coming from

The name entered circulation attached to a release window rather than a product. An analyst forecast relayed in late June put Z.ai's next flagship inside August 2026, and the numbering was assumed upward from there. August ended with GLM-5.3 shipping instead, and two things have since made "GLM-5.5" less likely rather than more.

The first is the vendor's own framing. On Z.ai's August 31 earnings call the next-generation base model was named as GLM-6.0 — a new generation, not a point release, and a number that skips 5.4 and 5.5 entirely. The second is cadence. Analysts reading Z.ai's release rhythm, roughly one flagship every 54 to 70 days, now put the next one in a window running from October 8 to November 9, 2026, and most of them expect it to be numbered GLM-5.4.

There was one genuinely fresh signal in the last week, and it is worth describing precisely because it is the kind of thing that becomes a false release story. On September 25, 2026 pre-registration strings reading "GLM 5.5 Flash" and "GLM-5.4" were spotted on a third-party coding tool's public data page — strings with no users behind them, no vendor confirmation, and no model attached. Re-checked on September 29, neither string appears on that page. That is a nomenclature sighting inside someone else's product, not a release, and it is the whole of the recent evidence.

The comparison you can actually run

Set Claude Sonnet 5.5 against GLM-5.3 instead and the matchup is real, current, and unusually interesting — but it is a different article from the one this page's title implies, so here is the short version rather than a rerun of a comparison we have already published.

• Price — Claude Sonnet 5.5 $2.00 / $10.00 per 1M vs GLM-5.3 $1.40 / $4.40 list, which our catalogue currently lists at $1.26 / $3.96

• Intelligence Index, v4.3.2 — Claude Sonnet 5.5 56 vs GLM-5.3 45, both at their maximum effort settings

• Cost per Index task — Claude Sonnet 5.5 $7.60 vs GLM-5.3 $2.01, a 3.8× gap against a 2.1× rate gap

• Output speed — Claude Sonnet 5.5 142 tokens/sec vs GLM-5.3 86.8 tokens/sec

• Verbosity on the Index — Claude Sonnet 5.5 410M output tokens vs GLM-5.3 210M

• Modality — Claude Sonnet 5.5 takes text, images and files; GLM-5.3 is text in, text out

• Weights — GLM-5.3 is downloadable from Z.ai's Hugging Face organisation under the vendor's own licence; Claude Sonnet 5.5 has none

The shape of that is familiar and the mechanism is the same one that runs through every Sonnet 5.5 matchup: the rate gap and the task gap differ because the token counts differ. A model that emits 410 million tokens where its rival emits 210 million turns a 2.1× price difference into a 3.8× bill. GLM-5.3's standing on the current revision is also worth stating plainly, because some older coverage still quotes it at 60 on a superseded Index revision: on v4.3.2 it scores 45, which places it second among open-weights models, behind MiMo-V2.6-Pro at 46 and just ahead of Kimi K3 at 44. If you have seen 60 attached to this model, that was a different board.

Both sides of a real pair on one key

The reason to hold an Anthropic model and a Z.ai model at the same time is that they fail differently, and running both normally costs two contracts. GLM-5.3 is in the OrcaRouter catalogue as z-ai/glm-5.3 at the provider rate with no markup added — one of 200-plus models behind a single OpenAI-compatible endpoint, with automatic failover between providers and a routing DSL for composing calls. Because the catalogue passes provider list price through unchanged, a Z.ai rate move lands on our side the same day it lands on theirs. Claude Sonnet 5 has been routable here since June 30 at Anthropic's $2 and $10, and remains the model most Claude API traffic runs on.

Claude Sonnet 5.5 is not in our catalogue yet, so the route to it today is Anthropic's own API — and that is also the honest limit of the routing advice here: the failover trick only helps once both models are somewhere you can reach them from one place.

What would have to appear

Turning "GLM-5.5" from a name into a model takes about a day of vendor work, and the evidence is easy to recognise when it lands. A model card with a parameter count and a context window. An API identifier on Z.ai's own model list. A price row. A Hugging Face repository under the Z.ai organisation, with a licence file, if the weights are open. And a page on Artificial Analysis, because a model nobody has measured is not a model you can compare. Until a majority of those five exist, any article that prints a GLM-5.5 spec sheet is printing somebody's projection.

The next checkpoint is the October 8 to November 9 window, and the number to watch for is 5.4 rather than 5.5. If Z.ai follows its own earnings call, the more interesting question is whether the next release is a point upgrade at all or the GLM-6.0 generation the company named. Either way, the comparison on this page will be rerun the week it happens — and it will be a comparison of two models with model cards.