Nanbeige4.3-33B
Open-weights 33B dense model for agentic coding, tool use, and long-context cowork tasks.
Updated 2026-10-26
Yes. Nanbeige4.3-33B is free to use.
Listed pricing: Free (Apache 2.0 open weights; ~66 GB BF16 or quantized).
Overview
Nanbeige4.3-33B is an open-weights dense 33B language model aimed at agentic coding, tool calling, and multi-step cowork workflows. Released as the teacher model for the Nanbeige4.3 series, it ships on Hugging Face under Apache 2.0 with a 512K context window—built for long repos, tool traces, and extended agent sessions rather than short chat replies.
It targets developers and open-weights teams who want a strong coding/agent backbone they can run locally or on multi-GPU setups (~66 GB BF16, or quantized for smaller footprints) without a closed API. Lab comparisons reported in launch coverage put it ahead of Qwen3.8-27B on most agentic benchmarks, which is the main reason it drew attention from the open-model community right after release.
What sets it apart from coding copilots like Cursor or Copilot is that it is a raw checkpoint, not a product UI: you wire it through your own stack (vLLM, llama.cpp-class runtimes, custom agents). That flexibility is the upside; the downside is you own serving, quantization choices, and evaluation—there is no managed IDE experience out of the box.
Is Nanbeige4.3-33B free?
Yes. Nanbeige4.3-33B is free to use.
What the free tier covers: Full model weights free under Apache 2.0; compute and hosting costs are on you.
Listed pricing: Free (Apache 2.0 open weights; ~66 GB BF16 or quantized).
Pricing on this page has not been re-verified. The entry was last edited on , and no separate pricing check has been run since. Treat the figures as a record of what was published then and confirm on the official site.
Nanbeige4.3-33B pricing
| Plan | Price | What's included |
|---|---|---|
| Open Weights | Free | Apache 2.0 download from Hugging Face; self-host BF16 (~66 GB) or quantized variants on local/multi-GPU hardware |
Apache 2.0 download from Hugging Face; self-host BF16 (~66 GB) or quantized variants on local/multi-GPU hardware
Pricing on this page has not been re-verified. The entry was last edited on , and no separate pricing check has been run since. Treat the figures as a record of what was published then and confirm on the official site.
Is Nanbeige4.3-33B worth it?
You can test that on the free tier before paying anything. The recorded trade-offs are listed below, and any one of them can settle the question on its own.
The 8.7/10 AI Score is an editorial read of published capability, price and shipping pace. Nobody here has hands-on hours with Nanbeige4.3-33B. How we verify.
Worth it if
The strengths recorded against this entry.
- Apache 2.0 open weights—no API fee or vendor lock for the checkpoint itself
- 512K context suited to long agent traces and large-repo workflows
- Positioned as teacher for Nanbeige4.3 with strong agentic-coding focus vs. general chat models
- Reported to beat Qwen3.8-27B on most agentic benchmarks in lab tests cited at launch
Not worth it if
Any one of these blocks your use case.
- ~66 GB BF16 footprint needs multi-GPU or heavy quantization for practical local use
- Raw model only—no IDE product, hosted chat, or managed agent UI
- Quality depends on your serving stack and prompt/tool harness; not a drop-in Copilot replacement
- Benchmark claims are from lab/release coverage; independent third-party evals may still catch up
Key features
512K Context
Supports up to 512K tokens so agent runs can keep large codebases, tool histories, and multi-file diffs in one session without aggressive truncation.
Agentic Coding
Tuned for multi-step coding agents—planning, editing, and verifying across files—rather than single-shot autocomplete alone.
Tool Use
Oriented toward structured tool calling so agents can invoke shells, search, and APIs as part of cowork-style workflows.
Open Weights
Apache 2.0 weights on Hugging Face let teams self-host, fine-tune, and redistribute without proprietary API lock-in.
How it compares
| Tool | Best for | Pricing | Score |
|---|---|---|---|
| Nanbeige4.3-33B | — | Free (Apache 2.0 open weights; ~66 GB BF16 or quantized) | 8.7/10 |
| Cursor | Professional developers handling complex, multi-file refactors who want AI built into a familiar VS Code-based editor. | Free Hobby + Individual $20/mo + Teams $40/user/mo + Enterprise Custom | 9.5/10 |
| GPT-5.5 | Developers and teams needing a frontier reasoning model for agentic coding workflows and large-codebase context handling. | API: $5/$30 per 1M tokens (in/out). ChatGPT Plus $20/mo, Pro $200/mo | 9.4/10 |
| Claude Code | Developers who want an agent that works inside an existing repo and toolchain rather than in a hosted editor, and who already pay for a Claude plan. | Included with Claude Pro $17/mo, Max 5x $100/mo, Max 20x $200/mo, Team and Enterprise plans | 9.3/10 |
Compare head-to-head
Comparison explorer Put Nanbeige4.3-33B up against any three tools Opens with Nanbeige4.3-33B already loaded. Add up to three more from the full index and read pricing, features, pros and cons in one table. →Related reading
Kimi K3.1 Pro vs Opus 5.5 on Scores and Access
Moonshot’s Kimi K3.1 Pro hits 65.7% on Terminal-Bench 4.0 against Claude Opus 5.5 at 66.4%, with live API access and open weights dated November 4.
Meta's $35B Class C Raise Buys Compute, Not Votes
Meta's October 2026 Class C equity raise funds AI data centers while preserving founder control and lifting the compute bar for smaller labs.
Brotli Unauthenticated RCE Claim, Audited
Security digests keep repeating Brotli RCE with “no account needed.” Here is what that claim measures, skips, and when it fails for your stack.
Ready to try Nanbeige4.3-33B?
Head to the official site to start with Nanbeige4.3-33B — pricing and plans are listed above.
Visit Nanbeige4.3-33B