Zixuan Li, head of Z.ai, introduces GLM-5.2, an open-weights model positioned near the frontier on long-horizon coding and agentic benchmarks. The talk covers the model's benchmark results relative to other frontier and open-weight models, the origin of the GLM name in a 2021 paper, why GLM is designed as more than a coding model, and Z.ai's rationale for releasing near-frontier weights openly: enabling on-premises deployment for enterprises and governments, domain fine-tuning (law, finance, security), and customer co-design. It closes with a demonstration of Z Code, Z.ai's coding harness built for GLM-5.2 and compatible with other frontier models. Timestamps included in the original record cover the intro, benchmarks, open-weights rationale, and Z Code sections. ## Video Identity Observations ### Observable Presentation - **Conference keynote format**: The video captures a remote keynote delivered at an AI industry event ("World's Fair"), introduced by a host: *"GLM is the talk of the town right now. There's been an amazing booth downstairs in the expo."* - **Live technical difficulties**: The speaker repeatedly could not see his own slides and relied on a co-presenter to advance them (*"since I cannot see my slides, so I will need Swee to help me flip through all the slides"*), giving the presentation an informal, improvised character. - **Product-launch framing**: The talk centers on introducing a new model release: *"it's the first time for us to introduce GLM 4.2 5.2 to the world."* ### Intended Audience - **AI practitioners and developers**, evidenced by references to coding tools (*"people use it inside clock code code X, open code"*), benchmarks (Terminal Bench, AIME), fine-tuning partners, and Hugging Face. - **The open-source community**, explicitly addressed: *"GLM 5.2 couldn't succeed without you, all the open source community players... you are the true hero."* - **Enterprise and government buyers**, particularly Western ones: *"if enterprise or government especially in the Western world, want to use the model, we open way, we upload to the hugging face so that they can use the model on premise."* - **Individual hobbyists/developers**, offered consumer-style access: *"we have a coding plan like the Codex or cloud core subscription for you to use your tokens as individuals."* ### Claimed Scope - **Near-frontier open-weights positioning** (matching the title): *"the capabilities are on par with at least Opus 4.7"* and *"actually leads the other open way model a lot and close to the frontier model."* - **Beyond coding**: The creator claims broader capability scope: *"GLM is more more than coding model... we improve a lot in GDP valve and also math problems... role play, general chat."* - **Company/model naming clarification**: *"...the company actually is called Zhipu... all the models, it's called GLM... GLM actually represent general language model pre-training with auto-regressive blank filling"* (2021 paper). A heritage claim is also made: *"we were one of the first labs to do explorations on large language models at the same time with OpenAI, and Anthropic, and DeepMind."* - **Open-weight rationale**: The creator articulates a three-part philosophy — user trust/on-premise needs, enabling domain fine-tuning diversity (*"Harvey is fine-tuning GLM 4.1"*), and co-shaping the model ecosystem's future. - **New tooling announcement**: The keynote's "one more thing" is a coding harness: *"Z code, actually it's built for Geo 5.2, but also support all frontier models. You can bring your own key."* ### Self-Description Evidence - The speaker positions Z.ai/GLM as an open-weights lab balancing business with ecosystem trust: *"we open the weights for several things because there are users' needs and there are our own needs."* - Community gratitude is a repeated stance: application builders and tool maintainers (*"Onflos and Media"*, Peter) are credited as partners in model success. - Direction of further engagement is stated: *"next time I'll definitely be in SF talking to everyone."* ### Distinguishing Claims vs. Demonstration - **Claims**: benchmark parity with frontier models, broad non-coding capabilities, open-source commitment — these are creator assertions backed only by slides referenced but not visible. - **Demonstrated**: the event context, remote delivery, audience composition (developers/open-source practitioners), and the speaker's transparent handling of the slide-control mishap are directly observable from the transcript. ### Limitations - No curated metadata was provided beyond the title. - The transcript contains numerous apparent transcription errors (e.g., "GLM 4.2/5.2/1.2/102/4.7" version numbers, "Geo 5.2," "clock code code X," "Kodak," "Onflos"), so specific tool and benchmark names should be treated as approximate. - Slide content itself is not visible; all benchmark claims rest on the speaker's narration. # Delivery Observation — "GLM-5.2: Open Weights, Near-Frontier Intelligence — Zixuan Li, Z.ai" ## Context of Delivery This is a **remote keynote** — the speaker presents over video link while a host flips slides on-site. The delivery is therefore shaped by a visible constraint: the speaker repeatedly cannot see his own slides. > "since I cannot see my slides, so I will need Swee to help me flip through all the slides for me" This produces a distinct, collaborative, slightly improvisational rhythm throughout the talk. --- ## Recurring Linguistic Patterns ### 1. Repetition for emphasis ("X and X") The speaker frequently doubles a word or phrase when stressing a point: > "we want to make the norm. We want to make my bets." > "there are users' needs and there are our own needs" > "you can ask the questions that he cannot answer in public" *(host segment)* > "like one of our like most prod uh proddest product and model" ### 2. Hedging and softening fillers Heavy use of **"like," "actually," "maybe," "I think,"** especially when introducing claims: > "Actually, it's not a a brand name, it's a generic term" > "I think it's a huge improvement for the open weight model" > "maybe some people have heard of it" > "I think the operation is is similar to Kodak" ### 3. Direct audience address ("you can…") A consistent invitation pattern — the speaker closes loops by telling listeners exactly what they can do next: > "you can just search for my name Zishan Lee and the Z A I org" > "you can try the inside the chatbot agent, and also you can call the API" > "you can just search Z code or you can go to our booth" ### 4. Anticipating objections He frequently voices the audience's skeptical question, then answers it: > "people always ask me, 'Why do you open weight? So, do you care about your business or like do you care about losing market to some infra suppliers?'" > "maybe people will think Z.AI and GLM they're irrelevant, right?" --- ## Teaching & Structuring Habits - **Enumerated reasoning**: The open-weights rationale is delivered as an explicit list — "several things" — then walked through with "**And second…**", "**And third…**", each anchored to a user need (security/control, diversity/fine-tuning, co-designing the future). - **Concrete naming of partners**: Claims are supported by real examples rather than abstractions: > "For example, Harvey is fine-tuning GLM 4.1, and maybe they're thinking about fine-tuning GLM 5.2 afterwards." - **Rhetorical framing of openness as shared credit**, ending the argument by crediting the community: > "I think uh you you are the true hero." - **Resource-forward close**: The talk ends by pointing at tangible artifacts (tech blog, Hugging Face repo, booth, coding plan) rather than summarizing. --- ## Handling of Live Friction (Observable Moments) These appear **isolated to this remote setup** and should not be read as stable habits: - Repeated slide coordination with the off-screen host: *"Yeah, so we're we're on the opening slide"*, *"Okay, the next slide"*, *"need to go back to the two two slides"*. - An off-script aside when the clicker lacks a back button: > "I would you I don't understand clickers that don't have back buttons. Like why Okay, I you know, anyway, go ahead." - Self-aware humor about presentation order: *"people want to see the GLM 5.2. They they don't want to see you like GLM 4.1 or 5."* --- ## Vocabulary Profile - **Benchmark and capability vocabulary dominates**: "intelligence upper bound," "token efficiency," "long horizon task," "thinking budget," "harness," "on premise." - **Community/ecosystem lexicon** clusters around the open-source argument: "open weight," "fine-tune," "co-shape the future," "ecosystem," "super developers." - Version numbers are spoken loosely and inconsistently in the transcript (GLM "4.2," "5.2," "4.7," "Opus 4.7 and 4.8"); whether this reflects the speaker's delivery or transcription error **cannot be determined from the transcript alone**. --- ## Uncertainty Notes - Only **one talk** is observed; habits like enumerated structure and objection-anticipation recur multiple times within this single delivery but cannot be confirmed as stable across appearances. - The host's closing segment (a second speaker) is included in the transcript; quotes from [737] onward belong to the host, not the keynote speaker, and were treated accordingly. - Some quoted phrases contain transcription artifacts ("proddest product"); these are reproduced as-is rather than corrected. # Observed Positions — GLM-5.2 Keynote (Z.ai) ## Video Metadata - **Title:** GLM-5.2: Open Weights, Near-Frontier Intelligence — Zixuan Li, Z.ai - **Note on speaker identification:** The transcript refers to the keynote speaker variously as "Zishan," "Zishan Lee," and (in the closing remarks) "Zhicheng," while the title names **Zixuan Li of Z.ai**. The transcript does not definitively reconcile these names; this record preserves them as expressed. --- ## Speaker 1: Moderator / Host (unnamed) ### Claims and remarks - Introduces the keynote speaker, noting the speaker has "been speaking with AI Singapore" and that **"GLM is the talk of the town right now."** - Notes the speaker is appearing remotely ("the magic of the internet") rather than in person, and that a Z.ai team is present at an expo booth. - Assists with slide navigation during the talk (the speaker states he cannot see his slides). - In closing remarks: - Empathizes with the remote situation: **"the very first World's Fair, I wasn't allowed to come back in the country for my own conference."** - States **four of eight** top Hugging Face contributors shown in the keynote's list are "present at World's Fair." - Says they are working with **Nvidia, Unsloth, Ollama** "and all those other fine-tuners" on inference and local tracks for the event. - Frames the World's Fair's purpose as bringing top AI companies together so attendees can "ask the questions that he cannot answer in public, but you can ask in private." - Teases upcoming sessions with **Hugging Face** and **Minimax**. --- ## Speaker 2: Keynote Speaker (Z.ai; named variously as Zishan / Zixuan Li / Zhicheng in the transcript) ### 1. Company and naming history - Apologizes for not attending in person; directs attendees to his X account and the Z.ai org account, and to the team booth. - The company is called **Zhipu**; the models are called **GLM**. - **GLM is not a brand name originally but a generic term**: *"GLM actually represent general language model pre-training with auto-regressive blank filling,"* from a paper published in **2021**. - Claims Zhipu was **"one of the first labs to do explorations on large language models at the same time with OpenAI, and Anthropic, and DeepMind."** - Qualification: they **no longer use the GLM architecture**, but retain "GLM" as the brand name for model versions. ### 2. Definition of intelligence - Argues intelligence is **not just IQ or benchmarks like AIME or physics problems**, pushing back on the framing that followed DeepSeek's and OpenAI's o1 launches. - From **GLM 4.5 to GLM 4.7**, the lab explored **reasoning, coding, and agentic capabilities**. ### 3. Coding and agentic capability claims - States **GLM 4.2 specializes in coding and agentic tasks** *(note: the transcript's version numbers are internally inconsistent — he references GLM 5.1, 5.2, 4.2, 4.1, 1.1, 1.2 at different points; these are preserved as spoken)*. - Addresses "rumors whether your model is close to mythos, fable" by citing slides: performance is **"somewhere between Opus 4.7 and 4.8."** - Cites benchmarks including **"Deep Sweep, Terminal Bench 211"** (noting Terminal Bench was "mentioned by the OpenAI team several minutes ago") and long-horizon tasks. - Claims capabilities are **"on par with at least Opus 4.7"** and show **significant improvement over 1.1**. - Introduces a new **"high" thinking level** for GLM 1.2 — the first time they've added a high thinking-budget tier — motivated by harder tasks consuming more tokens and their concern for **token efficiency**. - Claim: **"even without thinking, the non-thinking model is better than the 5.1 thinking model"** — framed as "a huge improvement for the open weight model" and the reason "people are talking about GLM 5.2 lately." ### 4. GLM is more than a coding model - Acknowledges people use GLM "inside clock code code X, open code" (apparently Claude Code / Codex-style tools), but says the model was trained broadly: - Improvements in **GDP valve** *(as transcribed)* and math ("we also care about math problems, frankly speaking"). - Training in **role play and general chat**. - Cites the **Artificial Analysis intelligence index**: GLM "leads the other open way model a lot and [is] close to the frontier model." - Encourages users to explore GLM "beyond the coding scope." ### 5. Rationale for open weights - Poses the audience's question: **"Why do you open weight? Do you care about your business or like do you care about losing market to some infra suppliers?"** - Gives three stated reasons: 1. **Users' needs + own needs**: users wanting "security and control" and trust-building; enterprise/government customers "especially in the Western world" can run the model **on premise** via Hugging Face. "Beneficial for the whole ecosystem… especially when the capabilities is close to the frontier model." 2. **Diversity**: domain fine-tuning in "legal, finance, security." Cites **Harvey fine-tuning GLM 4.1** and "maybe… thinking about fine-tuning GLM 5.2 afterwards"; says other companies see fine-tuning GLM as a differentiation strategy. 3. **Co-designing the future**: customers "need to see the architecture… the recipe of how you train the model." *"We want to make the norm. We want to make my bets. So, we want to co-shape the future with our customers with the open source community."* - Credits the open-source community (naming **Onflos and Media** *(as transcribed)*, individual super developers) and application builders ("like Peter"): **"you are the true hero."** - States GLM 5.2 "couldn't succeed without" the community, and that open source includes more than models — also open-source software and support. ### 6. Resources: tech blog and Z Code - Points to the **"Geo 5.2" tech blog** *(as transcribed)* containing: - Hugging Face repo links; chatbot agent and API access. - A **coding plan** ("like the Codex or cloud core subscription") for individual token use. - Details on the **training pipeline and recipe**, with the claim there is "fantastic technology behind the model, not just… a model that had great data." - **"One more thing" announcement**: first public release of their own harness, **Z Code**: - Built for GLM 5.2 but **supports all frontier models**; bring-your-own-key. - Claims operation "similar to Kodak" *(as transcribed, apparently Codex)*, supporting techniques like compaction "like what you've done in… Kodak and car code" *(as transcribed, apparently Claude Code)*. - Calls it "the perfect one for GLM" and invites attendees to the booth or to search "Z code." - Closes by promising to be **in San Francisco** next time. --- ## Unresolved tensions and qualifications (preserved as expressed) - The speaker **cannot see his slides**, leading to visible navigation confusion and partially unfinished slide commentary ("I haven't finished that slide"). - **Version numbering in the transcript is inconsistent** (GLM 4.2 / 5.2 / 1.2 / 4.1 / 5.1 / 1.1), and no clarification is available from the transcript alone. - Benchmark claims ("on par with at least Opus 4.7") are presented via slides not fully visible in the transcript and are framed as responses to rumors, not independently verified here. - The claim that Harvey "maybe" will fine-tune GLM 5.2 is explicitly tentative. - Speaker identity/naming discrepancy (title vs. transcript names) remains unverified from the source material. # Speaker & Provenance Record ## Video Identification - **Title:** *GLM-5.2: Open Weights, Near-Frontier Intelligence — Zixuan Li, Z.ai* - **Video ID:** `01a15ca7-26bd-7d49-bd67-016ad22186d2` - **Format:** Conference keynote delivered remotely at an AI "World's Fair" event, with an on-stage host handling slide advancement. ## Speakers & Roles | Speaker (as heard) | Role | Notes | |---|---|---| | **Zishan / Zishan Lee** (also rendered "Zhicheng" by the host; title metadata says "Zixuan Li, Z.ai") | **Keynote speaker, Z.AI representative** | Presented remotely ("magic of the internet"), could not see his own slides. Invited to connect via X: personal account "Zishan Lee" and company org "Z AI". | | **Swee** | **Host / emcee (AI Singapore affiliation)** | Introduced the keynote, noted prior speaking with AI Singapore, operated slides remotely, delivered closing remarks. | | **Ali** | Mentioned only | Referenced by host ("we'll talk to Hugging Face next… I might be doing Ali's job"); not heard on stage. | > **Name uncertainty flag:** The transcript and title give inconsistent spellings for the keynote speaker ("Zishan," "Zishan Lee," "Zhicheng," and title metadata "Zixuan Li"). Identity should be treated as one individual but spelling is unresolved. ## Speaker Attribution Boundaries - All claims about GLM models, Z.AI/Zhipu company history, benchmark positioning, and open-weight rationale come from the **keynote speaker**. - Closing remarks about the World's Fair, Hugging Face contributor presence, Nvidia/Unsloth/Ollama collaboration, and upcoming sessions (Hugging Face, MiniMax) come from the **host**. ## Cited / Referenced Sources (as claimed by speakers) - **GLM pre-training paper (2021):** described as "general language model pre-training with auto-regressive blank filling" — company identity: Zhipu. - **GLM-5.2 tech blog (referred to as "Geo 5.2" / possibly transcription artifact):** contains Hugging Face repo links, chatbot agent access, API access, coding-plan subscription, and training pipeline/recipe details. - **Benchmarks named by speaker:** DeepSweep, Terminal Bench 211 ("mentioned by the OpenAI team several minutes ago"), Artificial Analysis Intelligence Index, GDPval, AIME (as examples of IQ-style benchmarks). - **Comparison anchors claimed:** "somewhere between Opus 4.7 and 4.8," improvements over GLM 5.1/GLM 4.1 — third-party model claims stated by the speaker, not independently verified. - **Third parties named:** OpenAI, Anthropic, DeepMind, DeepSeek, Harvey (fine-tuning GLM 4.1), Unsloth, Ollama, Nvidia, Hugging Face, Peter (application builder), users of Claude Code/CodeX/OpenCode ecosystems. ## Promotional Segments (flagged) - **GLM-5.2 launch keynote** — core promotional framing: "first time for us to introduce GLM… to the world." - **Z Code harness** — explicitly promoted as the company's coding agent ("one more thing… our own harness"), with booth demos offered. - **Coding plan subscription** — marketed as analogous to Codex/Claude-style subscriptions. - **Expo booth invitation** — repeated encouragement to visit the Z.AI booth and team members. - **Host promotional framing** — World's Fair positioning ("bring all the top world's AI companies… so you can do business together") and teaser for upcoming Hugging Face and MiniMax sessions. ## Availability & Access Notes - Keynote delivered **remotely**; speaker stated a team is present at the expo booth. - Speaker promised future in-person appearance: *"next time I'll definitely be in SF."* - No transcript-recorded mention of a Q&A session; audience questions were directed to the booth or X contact. ## Unresolved / Uncertain Items - Speaker's exact name spelling (see flag above). - Several benchmark and product names appear as probable transcription artifacts (e.g., "GLM 4.2/5.2/1.1/1.2" used inconsistently; "Geo 5.2," "Deep Sweep," "clock code code X," "Kodak/car code" likely refer to GLM versions, the tech blog, DeepSweep, Claude Code/CodeX, and Codex/Claw Code respectively). Recorded as heard, not corrected. Without further ado, we're going to bring on Zishan for for his keynote if we're all set. Hey, how are you? \>\> Great I'm here. \>\> Yes, to the magic of the internet. It's really good to see you again. Zishan has you've been speaking with AI Singapore and obviously GLM is the talk of the town right now. There's been an amazing booth downstairs in the expo and I think a lot of people are just very excited to hear from you on what all is going on with ZIAI. So if you want to take it away. \>\> Yes, sorry again for not showing up in person, but I have a team whole team coming to the town and we have a booth. So you have any questions, feel free to reach out to me on X link and also you can look for my team on our booth. And since I cannot see my slides, so I will need Swee to help me flip through all the slides for me. Yeah. \>\> You're good. \>\> Here Swee, like what which slides we are in right now? \>\> Yeah, so we're we're on the opening slide. We're talking about intelligence and ZIAI. \>\> Okay, so yeah, so you can you can see that it's the first time for us to introduce GLM 4.2 5.2 to the world and also we are going to share something about Z.AI and GLM because maybe people will think Z.AI and GLM they're irrelevant, right? You are your company is not G.AI or your model's called Z1 or Z2. And you can find my X and our company's X account here, so you can just search for my name Zishan Lee and the Z A I org. And you can follow them for the follow-ups. So, yeah, we can go to the second slide. Yeah, actually I cannot see the slide, so I'll try to \>\> So, we we \>\> Yeah, try to make sure that it is correct. So, the company actually is called Zhipu. Maybe some people have heard of it. And the all the models, it's it's called GLM. Actually, it's not a a brand name, it's a generic term. So, GLM actually represent general language model pre-training with auto-regressive blank filling. And that paper was published back in 2021. So, actually we were the one of the first labs to do explorations on large language models at the same time with OpenAI, and Anthropic, and DeepMind. And even today, we we no longer use GLM as the architecture, we still use the name GLM as our brand name. So, we use GLM 5.1, 5.2, and it become like one of our like most prod uh proddest product and model. And the second thing that we look for is intelligence upper bound. So, in terms of intelligence, we we may feel that it's represent IQ, something something like that. And when Deep Seek launched one 01 launch, people are talking about the model's capability to solve math problems, physics problems. But what actually intelligence mean is not just like IQ or AIME or other other physics problem. So, from GLM 4.5 to GLM 4.7, we are exploring like several things, like reasoning, coding, and agentic capabilities. So, as you can see from the slides, So, we add like Yeah, the last slide. Yeah, we were actually I haven't like finished that slide. Yeah, so Okay. Yeah, need to go back to the two two slides. \>\> I don't have back I don't have a back button. Thank you. I would you I don't understand clickers that don't have back buttons. Like why Okay, I you know, anyway, go ahead. \>\> Yeah, never mind. Yeah, because people want to see the GLM 5.2. They they don't want to see you like GLM 4.1 or 5. But like GLM 4.2 actually specialize in coding and genetic task as you can see from the graph because there are a lot of rumors whether your model is close to mythos, fable, but actually I want to share these slides to all of you. So, you can see it's somewhere between Opus 4.7 and 4.8 and we use the hardest problems like Deep Sweep, Terminal Bench 211, which was mentioned by the Open AI team uh several minutes ago. And all the like long horizon task and benchmark shows that the capabilities are on par with at least Opus 4.7 and it it it shows a significant improvements over 1.1. Also, for GLM 1.2, we add a thinking uh level called high. So, because we also noticed as we move to the harder task it may consume more tokens and also we we care a lot about the token efficiency. So, it's the first time we add the high level for thinking budget. But even without thinking, the non-thinking model is better than the 5.1 thinking model. So, I think it's a huge improvement for the open weight model. That's what really impressed the world and why people are talking about GLM 5.2 lately. Okay, the next slide. And one one thing that I want to mention is that GLM is more more than coding model because people use it inside clock code code X, open code. But actually, we have trained a lot of things outside coding. For example, we improve a lot in GDP valve and also math problems. We also care about math problems, frankly speaking. And also, we train a lot of thing uh related to role play, general chat. We want to improve every aspect of the model. So, you can see from the artificial analysis intelligence index actually leads the other open way model a lot and close to the frontier model. So, I want you if you are you haven't experienced GLM yet, you can use GLM to do general chat, use it to process your daily workflow, not just for coding, but you can explore the model like beyond the coding scope. Next slide. And GLM 4.2 is a open weight model. So, people always ask me, "Why do you open weight? So, do you care about your business or like do you care about losing market to some infra suppliers?" But actually, we open the weights for several things because there are users' needs and there are our own needs. If we can meet their needs, I think it's okay. It's definitely okay for us to to open the model. For example, if our users want security and control and we want to build trust, we can open weight model. For for some cases, if enterprise or government especially in the Western world, want to use the model, we open way, we upload to the hugging face so that they can use the model on premise. I think it's very beneficial for the whole ecosystem it to explore the model, especially when the capabilities is close to the frontier model. And second, if people want diversity. So, in terms of diversity, I mean the capabilities in legal, finance, security, they can fine-tune the model. So, we need to open the model to let them fine-tune. For example, Harvey is fine-tuning GLM 4.1, and maybe they're thinking about fine-tuning GLM 5.2 afterwards. And I've I have like talked to a lot of other companies. They're also thinking about fine-tuning GLM as their like next step or next strategy to differentiate themselves from other application company. And third, if our customer or an individual want to co-design and predict the future, sometimes they need to see the architecture of the model. They need to see the recipe of how you train the model. So, we want to make the norm. We want to make my bets. So, we want to co-shape the future with our customers with the open source uh community. So, I think our needs and our ecosystem really like fits into each other. And GLM 5.2 couldn't succeed without you, all the open source community players like Onflos and Media and and some like individuals, super developers, they are part of it. And thanks a lot to application builders like Peter. Right? Because open source doesn't just include open source model, but also open source softwares and open source other sorts of support. So, all your support and what are you you're doing right now pushes to to make better models. I think uh you you are the true hero. And the last, I want to share a a great resource for you to go through. Geo 5.2 is our tech blog. So, actually in that tech blog, we share several things like our hugging face repo, how how you can try the Geo and 5.2. You can try the inside the chatbot agent, and also you can call the API. And also we have a coding plan like the Codex or cloud core subscription for you to use your tokens as individuals. And also we share something about our training pipeline training recipe, which you can understand why it's a good model. So, there are a lot of things behind the behind the model, not just um a a model that had great data. We also have fantastic technologies behind the model, so you can export the model yourself. You can see what difficulties we have gone through. And the last slide, actually it's kind of a one more thing. So, it's the first time we share the code to the whole community. Woo! So, that one more thing is we actually have our own harness. The Z code, actually it's built for Geo 5.2, but also support all frontier models. You can bring your own key. You can connect to Z code. Actually, the I think the operation is is similar to Kodak. Actually, you you can try some techniques like go or other like compact technique techniques like what you've done in in Kodak and car code. And this harness is I think it's the perfect one for GLM 102. If you haven't experienced it, you can just search Z code or you can go to our booth. We have our team members showing this harness to you. And welcome to to our booth. And welcome to like talk to me in the future. And next time I'll definitely be in SF talking to everyone. Yeah, thanks. \>\> Thank you very much. Um So uh trust me, the very first World's Fair, I wasn't allowed to come back in the country for my own conference, so I know exactly this feeling. Uh but it it it's all good. Uh the ZAI team, I really appreciate them uh making the effort. They really want to meet you. They're here to meet you. And this is the whole point of the World's Fair, to bring all the top world's AI uh companies and labs uh all in one place, so you can do business together, uh meet the people behind the models that you use, ask the questions that he cannot answer in public, but you can ask in private. Um I'm also very proud. Uh he showed uh Zhicheng showed that that list of uh Hugging Face uh you know, top uh contributors. I think we are four for eight uh in that list uh present at World's Fair. Uh we're working with Nvidia and Unsloth and Ollama and then uh all those other fine-tuners as well to uh to make our our sort of the inference and the local tracks uh that you're going to see over the next few days. Um with that, thank you so much to Zhicheng. Uh I'm going to invite on uh the next uh couple folks. I think I I I think I might be doing Ali's job here, so uh we'll we'll talk to Hugging Face next and Minimax. Thank you.