Does Sume offer Haiku 5.5, GLM 5.3 Flash or Mistral Large 4?
Checked against Sume's model catalog on 2026-10-08: Haiku 5.5 and GLM 5.3 Flash have enabled rows; Mistral Large 4, Clef and Strands Decider 2B have none.

As of 2026-10-08, Sume's agent model catalog in the repository has enabled rows for Claude Haiku 5.5 and GLM 5.3 Flash. It has no row for Mistral Large 4, Cloudflare Clef, Amazon Strands Decider 2B, or a model named GLM 5.3 Fast. Separately, the public Agent Completions model field accepts only sume-agent.
The answer, model by model
I checked the catalog file that defines every model id the product can name, and compared each against the vendor's own page. A catalog row means the product can name the model in its Agents model picker; which environment shows it depends on gates, so confirm in your own workspace.
| Model | Vendor fact (page read 2026-10-08) | In Sume's agent catalog? |
|---|---|---|
| Claude Haiku 5.5 | Released Oct 7; $0.10 / $0.50 per million tokens up to 100K prompt (Anthropic) | Yes, enabled row |
| GLM 5.3 Flash | 1M context, 128K max output, function calling, JSON output (Z.ai docs) | Yes, enabled row |
| GLM 5.3 Fast | No vendor page fetched | No row found |
| Mistral Large 4 | Upcoming; ETA Oct 31 (Hugging Face model page) | No row found |
| Cloudflare Clef | 27B decision model, Apache-2.0 (Cloudflare pages) | No row found |
| Strands Decider 2B | Decision model, Apache-2.0 (Hugging Face card) | No row found |
Two different questions
There are two surfaces, and people mix them. The Agents chat UI and Scheduled automations pick a model. The developer API for Agent Completions does not: its model field accepts only sume-agent, and omitting it gives you the same agent. A value other than sume-agent returns 400 invalid_request.
The Scheduled docs say a schedule has "a model" next to its cron expression and spend cap, and that you choose it when you write the instructions in the dashboard.
What I dropped
The brief for this week named GLM 5.3 Fast. I found no Sume catalog row by that name, and the only vendor page I could read describes GLM-5.3-Flash and a FlashX variant. So this post makes no claim about a Fast model. I also make no claim about prices of GLM or Mistral, because I did not read a vendor price page for them.
How to use this
The points that matter here, in the order you will hit them:
- For the developer API, send
sume-agentor omit the field. - For Agents chat and schedules, open the picker and use what it shows.
- Run decision models such as Clef or Strands Decider in your own service, before you call Sume.
- Re-check after a release; a name in the news is not a name in a catalog.
How the catalog was checked
The catalog declares each model id once, with a label, an enabled flag, a gate that says which environment lists it, and a route. Claude Haiku 5.5 and GLM 5.3 Flash have enabled: true and an OpenRouter gate. The older Haiku 4.5 row is disabled and points to Haiku 5.5 as its successor, so stored threads that used it move forward. No row exists for the other names in the table. I did not verify pricing in Sume for any of these models, so this post gives vendor prices only where a vendor page was read.
Sources
- Anthropic, Claude Haiku page (read 2026-10-08)
- Z.ai docs, GLM-5.3-Flash (read 2026-10-08)
- Hugging Face, Mistral-Large-4.0-1T05-A52B model page (read 2026-10-08)
- Cloudflare Workers AI docs, Clef (read 2026-10-08)
- Hugging Face, strands-decider-2B-hobson-v19 model card (read 2026-10-08)
- Sume docs, Agent Completions
Related posts
More in Models
- Does SynthID survive captions and a 9:16 crop on Omni clips?
Google says SynthID is built to survive cropping, filters and compression. What that means for a captioned or reframed Omni clip, and what Sume claims.
- Edit a clip, swap a person, or move motion onto photos: 3 Sume models
Omni Flash edits a clip by prompt ($1.25 per 10 s, 720p), H3 Max Recast swaps people ($3.75, 768p), Genjutsu moves motion onto photos ($3.98, 480p).
- Edit with input_references: 5 of 19 Sume image models are text-only
Five Sume image models cannot take input_references: Soul, Imagen 4 Fast, Imagen 4 Ultra, Recraft V4 and Qwen Image Max. Here are the 14 that can, with prices.
- EU, UK or South Korea team: which open video weights you may use
HunyuanVideo and MiniMax H3 licenses exclude the EU, UK and South Korea. Wan 2.2 is Apache 2.0 and LTX-2.5 is worldwide with a revenue line.
Written by Sume