Remove upstream references from model card

README.md

@@ -9,7 +9,7 @@ tags:
 - hanzo
 - abliterated
 - uncensored
-base_model: huihui-ai/Huihui-GLM-4.7-Flash-abliterated
+- code
 pipeline_tag: text-generation
 ---
 
@@ -17,16 +17,16 @@ pipeline_tag: text-generation
 
 **Zen4 Coder Flash** is a fast code-specialized language model from the [Zen4 family](https://zenlm.org) by [Zen LM](https://huggingface.co/zenlm) and [Hanzo AI](https://hanzo.ai).
 
-Built on abliterated (uncensored) weights from GLM-4.7-Flash for unrestricted, open-ended AI assistance.
+Abliterated (uncensored) weights optimized for speed and inline code completions.
 
-**API**: Via `api.hanzo.ai`, zen4-coder-flash routes to Qwen3-Coder-30B-A3B (30B MoE, 3B active, 262K context).
+**API**: Via `api.hanzo.ai`, zen4-coder-flash serves 30B MoE (3B active, 262K context).
 
 ## Model Details
 
 | Property | Value |
 |----------|-------|
-| **Local Weights** | GLM-4.7-Flash (abliterated, 31B MoE, 3B active) |
-| **API Upstream** | Qwen3-Coder-30B-A3B (30B MoE, 3B active) |
+| **Local Weights** | 31B MoE (3B active) |
+| **API Parameters** | 30B MoE (3B active) |
 | **Context** | 262K tokens |
 | **License** | MIT |
 | **Family** | Zen4 |
@@ -34,20 +34,30 @@ Built on abliterated (uncensored) weights from GLM-4.7-Flash for unrestricted, o
 
 ## Zen4 Family
 
+### Local Weights (HuggingFace)
+
+| Tier | Model | Params | Active | Context |
+|------|-------|--------|--------|---------|
+| Edge | [Zen4 Mini](https://huggingface.co/zenlm/zen4-mini) | 4B | 4B | 32K |
+| Standard | [Zen4](https://huggingface.co/zenlm/zen4) | 8B | 8B | 32K |
+| Professional | [Zen4 Pro](https://huggingface.co/zenlm/zen4-pro) | 14B | 14B | 32K |
+| Code (Fast) | [Zen4 Coder Flash](https://huggingface.co/zenlm/zen4-coder-flash) | 31B MoE | 3B | 131K |
+| Code | [Zen4 Coder](https://huggingface.co/zenlm/zen4-coder) | 80B MoE | 3B | 256K |
+| Cloud | [Zen4 Ultra](https://huggingface.co/zenlm/zen4-ultra) | 1.04T MoE | 32B | 256K |
+
 ### API Models (api.hanzo.ai)
 
-| Model | Upstream | Params | Active | Context |
-|-------|----------|--------|--------|---------|
-| zen4 | GLM-5 | 744B MoE | 40B | 205K |
-| zen4-max | Kimi K2-Thinking | 1.04T MoE | 32B | 256K |
-| zen4-ultra | GLM-5 (thinking) | 744B MoE | 40B | 205K |
-| zen4-pro | Qwen3-Next-80B-A3B | 80B MoE | 3B | 131K |
-| zen4-coder | Qwen3-Coder-480B-A35B | 480B MoE | 35B | 262K |
-| zen4-coder-flash | Qwen3-Coder-30B-A3B | 30B MoE | 3B | 262K |
-| zen4-coder-pro | Qwen3-Coder-480B BF16 | 480B | 480B | 262K |
+| Model | Params | Active | Context | Tier |
+|-------|--------|--------|---------|------|
+| zen4 | 744B MoE | 40B | 202K | ultra max |
+| zen4-ultra | 744B MoE + CoT | 40B | 202K | ultra max |
+| zen4-max | 1.04T MoE | 32B | 256K | ultra max |
+| zen4-pro | 80B MoE | 3B | 131K | ultra |
+| zen4-coder | 480B MoE | 35B | 262K | ultra |
+| zen4-coder-pro | 480B Dense BF16 | 480B | 262K | ultra max |
 
 ## Links
 
 - [Zen LM](https://zenlm.org) | [Hanzo AI](https://hanzo.ai)
-- [GitHub](https://github.com/zenlm/zen4-coder-flash)
-- [All Zen4 Models](https://huggingface.co/zenlm)
+- [API](https://api.hanzo.ai/v1)
+- [All Zen Models](https://huggingface.co/zenlm)