Nanbeige4.3-33B
Nanbeige4.3-33B by Nanbeige is an Apache 2.0 open-weight 33B teacher model for agentic coding and cowork tasks, with 512K context and configurable thinking.
Model Overview
Capabilities, design details, and architectural traits
Nanbeige4.3-33B - Teacher Agent for the Nanbeige4.3 Series
Nanbeige4.3-33B is a dense 33B-parameter open-weight model from Nanbeige, built for agentic coding and professional cowork tasks rather than general chat. Its defining role is unusual: it is the flagship teacher model of the Nanbeige4.3 series, intended to anchor a distillation pipeline that carries advanced agentic capabilities into smaller models such as a planned Nanbeige4.3-3B.
| Trait | Detail |
|---|---|
| Teacher model role | Designed as the high-capacity teacher agent that will anchor distillation of agentic capabilities into smaller Nanbeige4.3 models, including a planned 3B variant. |
| Dual agentic focus | Targets two documented categories: agentic coding (terminal and repository work, multi-step software engineering) and agentic cowork (non-coding professional knowledge work across many occupations). |
| 512K context window | Supports up to 524,288 tokens, suited to large codebases and long multi-step agent workflows. |
| Configurable chat template | Ships with enable_thinking and preserve_thinking flags for reasoning control, plus a tool-use template where tool_call_format="xml" is recommended for best tool-calling performance. |
| Open Llama architecture | Declares LlamaForCausalLM with 52 layers, GQA attention and a 166,912-token vocabulary, released under Apache 2.0. |
| Serving options | Deployable through Transformers, SGLang, or vLLM, with documented tensor-parallel serving configurations. |
Thinking and tool-use behavior
The tokenizer exposes two reasoning switches. enable_thinking turns reasoning generation on or off per response. preserve_thinking keeps reasoning from earlier assistant turns in multi-turn conversations, recommended as True for multi-turn tool use, office tasks, and code-agent workflows, and False for general chat.
Distillation-first design
Nanbeige frames Nanbeige4.3-33B as a practical deployment model and as the foundation of its small-model distillation strategy. The team invites collaboration on domain-specific and industry-focused agents, including joint work on data and training.
Benchmark Performance
Independent evaluations · Artificial Analysis
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Nanbeige | Anthropic | OpenAI |
| Release Date | October 25, 2026 | November 9, 2026 | November 17, 2026 |
| Knowledge Cutoff | - | - | - |
| Context & Limits | |||
| Context Window | 524K | 1M | 1.1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | Free Best Input Pricing | $8 | $8 |
| Output Pricing | Free Best Output Pricing | $40 | $40 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 20.4 | 64.6 Best Intelligence Index | 62.1 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Benchmark comparison not available
Select at least two models that have available benchmark evaluation data to render the performance comparison charts.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Top AI Models
Leading alternatives by intelligence score