This 33B Model Was Built to Be a TEACHER (Nanbeige 4.3 beats Qwen at agents)
This video explains how the Apache-2.0 Nanbeige 4.3-33B serves as an agent-focused teacher model whose capabilities can be distilled into smaller students, while comparing its reported strengths and losses across coding, terminal, co-work, and knowledge benchmarks. It also translates the model's 66 GB half-precision size and 512K context window into practical Q4, Q8, MLX, Ollama, and agent-harness choices for self-hosting.
Scottek7 minTranscript found
Quick learning frame
Read this before watching.
A local runtime lesson is about fit: model, quantization, hardware, endpoint, latency, privacy, tool integration, and task limits.
New playlist item from Scottek; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Skill you build: The ability to decide whether an open agent model fits a workload and local machine by matching benchmark type, quantization, memory needs, context size, thinking mode, and tool-call format.
Watch for the shift from claim to mechanism. The learning value is the point where the transcript reveals a repeatable action, tool boundary, context move, review habit, or artifact.
Concept diagram
Where this video fits.
01Task
02Hardware
03Model/quantization
04Runtime endpoint
05Agent tool loop
06Benchmark task
07Fallback
Deep lesson
Turn this video into working knowledge.
944 cleaned transcript words reviewed across 324 timed caption segments.
Thesis
This 33B Model Was Built to Be a TEACHER (Nanbeige 4.3 beats Qwen at agents) teaches a practical local model/runtime move: This video explains how the Apache-2.0 Nanbeige 4.3-33B serves as an agent-focused teacher model whose capabilities can be distilled into smaller students, while comparing its reported strengths and losses across coding, terminal, co-work, and knowledge benchmarks. It also translates the model's 66 GB half-precision size and 512K context window into practical Q4, Q8, MLX, Ollama, and agent-harness choices for self-hosting.
The goal is not to remember the video. The goal is to extract the operating principle, tie it to timestamped evidence, test how far the claim transfers, and make something reusable.
0:22
Teacher Before Student
“doesn't just teach. On agentic coding benchmarks, it leads its entire weight class. Terminal bench, deep swe, a full bench of co-work tasks, ahead of Qwen, ahead of Gemma, ahead of Anthropic's small model. It's Apache 2.0. It...”
Nanbeige 4.3-33B is a dense 52-layer Llama-architecture teacher built around agentic coding and professional co-work rather than chat scores; its skills are intended for distillation into smaller models, including an announced 3B student. It uses thinking by default, supports native tools with XML preferred over JSON, and offers a 512,000-token context window under Apache 2.0. Write a four-line model profile listing Nanbeige's teacher role, target workloads, default reasoning behavior, and preferred tool-call format.
3:07
Choose the Game
“everything. SW E Bench Pro goes to Qwen, 61.7 to 60.3. GPQA Diamond goes to Gemma. GDP Eval goes to Qwen. And these are the team's own reported numbers. Same harness, highest thinking effort. Vendor benchmarks flatter by...”
The team's reported results show large leads on agentic tasks such as Terminal Bench 2.1, Terminal Bench 4, Automation Bench, and Job Bench, but Qwen wins SW E Bench Pro and GDP Eval while Gemma wins GPQA Diamond. Because these are vendor-reported numbers, the useful signal is the pattern: Nanbeige is strongest for tools, terminals, and plans, not universally best at knowledge recall. Sort the named benchmarks into agentic work and knowledge or evaluation work, then note which model wins each reported exception before choosing a workload for Nanbeige.
5:44
Quantize to Fit
“else. First, the thinking toggle. Leave it on for agents, code, and tool workflows. Switch it off for plain chat and watch your tokens per second roughly double with barely any quality loss on simple tasks. Second, the...”
Full half-precision weights require 66 GB, while a Q4 build fits comfortably on one 24 GB GPU; Q8 needs about 35 GB, and a 4-bit MLX build was reported using 19 GB at short context and 48 GB at 128K on a 64 GB Mac. Start around 32K context because memory use grows with filled context, keep thinking on for agent and tool work, turn it off for simple chat, and configure XML tool calls in the harness. Choose a configuration for either a 24 GB GPU or a 64 GB Mac, specifying quantization, starting context, thinking mode, and tool-call dialect.
01
Task
Start with this video's job: This video explains how the Apache-2.0 Nanbeige 4.3-33B serves as an agent-focused teacher model whose capabilities can be distilled into smaller students, while comparing its reported strengths and losses across coding, terminal, co-work, and knowledge benchmarks. It also translates the model's 66 GB half-precision size and 512K context window into practical Q4, Q8, MLX, Ollama, and agent-harness choices for self-hosting. Treat "Task" as the outcome you are trying to make visible, not a topic label. Anchor it to 0:22, where the video says: “doesn't just teach. On agentic coding benchmarks, it leads its entire weight class. Terminal bench, deep swe, a full bench of co-work tasks, ahead of Qwen, ahead of Gemma, ahead of Anthropic's small model. It's Apache 2.0. It...”
02
Hardware
Use "Hardware" to locate the part of the local model/runtime mechanism the video is demonstrating. Ask what changes in your real setup if this claim is true. Anchor it to 3:07, where the video says: “everything. SW E Bench Pro goes to Qwen, 61.7 to 60.3. GPQA Diamond goes to Gemma. GDP Eval goes to Qwen. And these are the team's own reported numbers. Same harness, highest thinking effort. Vendor benchmarks flatter by...”
03
Model/quantization
Turn "Model/quantization" into the reusable artifact for this lesson: A local model/runtime fit sheet with hardware constraints, model choice, endpoint setup, task benchmark, and fallback rule. This is where watching becomes something you can inspect and reuse.
04
Runtime endpoint
Use "Runtime endpoint" as the application surface. Decide whether the idea touches a browser flow, a local file, a model choice, a source document, a UI, or a review step.
05
Agent tool loop
Use "Agent tool loop" to prove the lesson. The evidence should connect back to the video title, transcript anchors, and a concrete output, not a generic best-practice claim.
06
Benchmark task
Use "Benchmark task" to carry the idea forward: save the prompt, checklist, diagram, or operating rule that would make the next agent run better.
07
Fallback
Connect "Fallback" to This 33B Model Was Built to Be a TEACHER (Nanbeige 4.3 beats Qwen at agents) by naming the claim, the evidence, and the artifact it should produce.
Example
Source-backed artifact packet
Convert the video into a scoped artifact request that includes the transcript claim, mechanism, acceptance criteria, and proof. The output should be a local model/runtime fit sheet with hardware constraints, model choice, endpoint setup, task benchmark, and fallback rule..
Example
Local model/runtime proof brief
Separate what the speaker claims, what the demo actually proves, and what still needs outside verification before you adopt the local model/runtime pattern.
Example
Teach-back module
Transform the lesson into a definition, a Task -> Hardware -> Model/quantization -> Runtime endpoint -> Agent tool loop -> Benchmark task -> Fallback diagram, one misconception, one practice exercise, and a check-for-understanding question.
Do not learn it wrong
Treating the title as the lesson without checking what the transcript actually says.
using a local model like ChatGPT
ignoring latency/context limits
no benchmark task
Letting the lesson drift into local-model ideology.
Letting the lesson drift into hardware specs without workflow fit.
Letting the lesson drift into benchmarks unrelated to the actual task.
Do not count this as learned until these are true.
01
State the transcript-backed claim in your own words: This video explains how the Apache-2.0 Nanbeige 4.3-33B serves as an agent-focused teacher model whose capabilities can be distilled into smaller students, while comparing its reported strengths and losses across coding, terminal, co-work, and knowledge benchmarks. It also translates the model's 66 GB half-precision size and 512K context window into practical Q4, Q8, MLX, Ollama, and agent-harness choices for self-hosting.
02
Explain the practical stakes without hype: New playlist item from Scottek; queued for transcript-backed review, topic mapping, and a practical learning artifact.
03
Map the idea onto the Task -> Hardware -> Model/quantization -> Runtime endpoint -> Agent tool loop -> Benchmark task -> Fallback sequence and name the weakest link.
04
Produce the artifact and include the evidence that proves it: A local model/runtime fit sheet with hardware constraints, model choice, endpoint setup, task benchmark, and fallback rule.
Put it into practice
Give this grounded prompt to Codex or Claude after watching.
You are helping me turn one specific YouTube video into real, durable learning.
Source video:
- Title: This 33B Model Was Built to Be a TEACHER (Nanbeige 4.3 beats Qwen at agents)
- URL: https://www.youtube.com/watch?v=a4vtgwm5vaM
- Topic: Agent Architecture
- My current learning frame: Given one local machine and one intended workload, write a Nanbeige deployment card that selects a quantization, starting context, thinking setting, tool format, and benchmark evidence supporting the choice.
- Why this matters: New playlist item from Scottek; queued for transcript-backed review, topic mapping, and a practical learning artifact.
Transcript anchors from this exact video:
- 0:22 / Evidence 1: "doesn't just teach. On agentic coding benchmarks, it leads its entire weight class. Terminal bench, deep swe, a full bench of co-work tasks, ahead of Qwen, ahead of Gemma, ahead of Anthropic's small model. It's Apache 2.0. It..."
- 3:07 / Evidence 2: "everything. SW E Bench Pro goes to Qwen, 61.7 to 60.3. GPQA Diamond goes to Gemma. GDP Eval goes to Qwen. And these are the team's own reported numbers. Same harness, highest thinking effort. Vendor benchmarks flatter by..."
- 5:44 / Evidence 3: "else. First, the thinking toggle. Leave it on for agents, code, and tool workflows. Switch it off for plain chat and watch your tokens per second roughly double with barely any quality loss on simple tasks. Second, the..."
- 7:19 / Evidence 4: "want the 3B the moment it lands, subscribe. I'll bench it the same week. Links below. Synthetic narration, real numbers."
Video-aware target:
- Prompt lane: Local model/runtime
- Mechanism to extract: Identify why the local setup works or fails for this specific agent task, not whether local models are generally good.
- Artifact to produce: A local model/runtime fit sheet with hardware constraints, model choice, endpoint setup, task benchmark, and fallback rule.
- Artifact must include: hardware; runtime; model/quantization; endpoint; agent integration; benchmark/fallback
Your task:
1. Use the transcript anchors above as the primary source packet. If you add outside context, label it clearly as outside context and keep it secondary.
2. Create a source-check table with columns: timestamp, claim, transcript support, what the demo proves, confidence, and what still needs verification.
3. Extract the actual teachable mechanism from the video: Identify why the local setup works or fails for this specific agent task, not whether local models are generally good. Do not invent claims that are not supported by the title, lesson frame, or transcript anchors.
4. Build a reusable learning artifact: A local model/runtime fit sheet with hardware constraints, model choice, endpoint setup, task benchmark, and fallback rule.
5. Include:
- a plain-English definition of the core idea
- a diagram or structured model using this sequence: Task -> Hardware -> Model/quantization -> Runtime endpoint -> Agent tool loop -> Benchmark task -> Fallback
- answers to these source questions: What machine/runtime is shown? | What task exposes the model limit? | What setup change improves the loop?
- 3 concrete examples that apply the video idea to real agentic work, such as Ollama or LM Studio coding endpoint; MLX Apple Silicon runner; DGX-backed Hermes session
- 2 failure modes the video helps prevent, chosen from the transcript evidence and these likely risks: using a local model like ChatGPT; ignoring latency/context limits; no benchmark task
- a checklist for the next real workflow, focused on: task fit, runtime setup, latency/context, tool loop, fallback
- one practical exercise with a clear done signal: Choose one real coding task and specify the pass/fail benchmark for a local model.
6. Add a "learning transfer" section: what changes in my workflow tomorrow if I actually learned this?
7. Add a "source check" section that cites which transcript anchor supports each major takeaway.
Quality bar:
- Make this specific to "This 33B Model Was Built to Be a TEACHER (Nanbeige 4.3 beats Qwen at agents)", not a generic Agent Architecture essay.
- Tie each harness element to a transcript anchor that names a tool, state boundary, permission, model behavior, or verification step.
- Prefer operational examples, failure modes, and reusable artifacts over broad definitions.
- Call out uncertainty instead of smoothing over weak evidence.
- Avoid these generic drifts: local-model ideology; hardware specs without workflow fit; benchmarks unrelated to the actual task.
- If evidence is weak or missing, stop and say what transcript segment or timestamp needs review instead of guessing.
- Finish with a concise artifact I could paste into my learning app.
Misconceptions
What to stop believing.
A better model automatically makes a better agent.
The model matters, but harness design determines whether the system can act safely and repeatably.
More tools always help.
Every tool increases surface area. Strong agents have the right tools with clear permissions.
Memory means saving everything.
Useful memory is compressed, curated, and tied to future decisions.
Practice studio
Learning only counts when you make something.
01
Transcript evidence map
Separate what the video actually says from what you already believe about the topic.
3 source-backed takeaways with timestamps, confidence, and a transfer note.02
One useful artifact
Apply the video to a real workflow and produce a local model/runtime fit sheet with hardware constraints, model choice, endpoint setup, task benchmark, and fallback rule..
A reusable artifact with a done signal and one verification step.03
Local model/runtime teach-back card
Explain the local model/runtime mechanism to someone who has not watched the video yet.
A 90-second explanation, one diagram, one example, and one misconception to avoid.
Recall check
Answer first, then reveal — without rewatching.
Why is Nanbeige 4.3-33B described as a teacher model?
What qualification does the video place on Nanbeige's benchmark leadership?
What configuration does the video recommend for running Nanbeige on one 24 GB Nvidia GPU?
Source shelf
Use the video as a doorway, then verify with primary sources.
