feature Qwen3.8-27B and K2.7-Code; add measured-and-corrected section; surface our most-downloaded model

README.md

@@ -13,7 +13,8 @@ colorTo: purple
   <a href="https://cis.bati.ai"><img src="https://img.shields.io/badge/Enterprise-Bati%20CIS-violet?style=for-the-badge"></a>
 </p>
 
-> **Kimi K2.6 — 1T MoE, SWE-Bench Pro 58.6** (beats GPT-5.4 xhigh, Claude Opus 4.6) — running locally on M3 Ultra.
+> **Qwen3.8-27B — quantized 2 days after the weights dropped**, six sizes, vision, Korean-verified.
+> **Kimi K2.7-Code — 1T MoE**, +21.8% coding over K2.6 — running locally on M3 Ultra.
 > **Gemma 4 E4B — 57 tokens/sec** on a 16GB Mac mini M4.
 > **No API costs. No rate limits. No cloud.**
 
@@ -27,8 +28,11 @@ We quantize popular open-weight models for **every Mac** — direct from officia
 # 16GB Mac mini M4 — entry, 57 t/s
 ollama pull batiai/gemma4-e4b:q4
 
+# 32GB+ Mac — newest 27B, vision + 262K context
+ollama pull batiai/qwen3.8-27b:iq4
+
 # 512GB M3 Ultra — 1T MoE frontier
-ollama pull batiai/kimi-k2.6:iq4
+ollama pull batiai/kimi-k2.7-code:iq4
 ```
 
 ---
@@ -43,6 +47,8 @@ Every speed measured on actual hardware. Full reports in each model card. We con
 | MacBook Air 16GB | `batiai/qwen3.5-9b:q4` | 5.2 GB | 12.5 t/s | Tool calling, JSON |
 | Mac mini M4 Pro 24GB | `batiai/gemma4-26b:iq4` | 15 GB | **85 t/s** | MoE, larger context |
 | MacBook Pro 32GB | `batiai/nemotron3-nano:iq4` | 17 GB | — | NVIDIA NemotronH MoE, hybrid Mamba+Attention |
+| **MacBook Pro 32GB+** | [`batiai/qwen3.8-27b:iq4`](https://huggingface.co/batiai/Qwen3.8-27B-GGUF) | 15 GB | **15.5 t/s** | **Newest 27B** — vision, 262K context, thinking |
+| MacBook Pro 24GB+ | [`batiai/qwen3.6-27b:iq4`](https://huggingface.co/batiai/Qwen3.6-27B-GGUF) | 15 GB | — | Our most-downloaded model (150K+) |
 | MacBook Pro 48GB | `batiai/qwen3.6-35b:iq4` | 22 GB | ~30 t/s | Tools + thinking, MoE |
 | MacBook Pro 128GB | `batiai/nemotron3-super:iq4` | 62 GB | — | NemotronH 120B-A12B, agentic |
 | MacBook Pro 96GB | `batiai/qwen3.6-35b:q6` | 29 GB | ~27 t/s | Top quality chat |
@@ -83,8 +89,10 @@ Quantized from the **publisher's official FP8/BF16 weights** — never re-quanti
 </td>
 <td width="50%" valign="top">
 
-### Verified on Real Macs
-Tested on **Mac mini M4 16GB** + **MacBook Pro M4 Max 128GB**. Korean validation, tool-call JSON, 200-token throughput — measured, reproducible, documented in each model card.
+### Verified on Real Macs — and corrected in public
+Tested on **Mac mini M4 16GB** + **MacBook Pro M4 Max 128GB**. Korean validation, tool-call JSON, throughput — measured, reproducible, documented in each card.
+
+When measurement contradicts us, we change the card. On Qwen3.8-27B we advertised 16 GB targets, measured it, found the model doesn't fit, and said so. We also found **`Q3_K_M` is smaller *and* slower than `IQ4_XS` on Apple Silicon** — reproducing across four model families — and moved our recommendation. Both are written up in the card, including what we got wrong.
 
 </td>
 </tr>