@bendyline/gilde 0.1.30 → 0.1.32

package/authoring/chat-models/btl4-35b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "btl4-35b-q4",
   "name": "BTL-4 (35B-A3B, Q4)",
-  "description": "Bad Theory Labs' agentic fine-tune of Ornith 1.0 35B, at 35B total parameters with ~2B active per token (Mixture-of-Experts: 256 experts, 8 routed). 30 of its 40 layers use linear attention, so the KV cache costs only ~20 KB/token and the full 256K context stays affordable. Q4_K_M quantization from bartowski — the full-quality companion to the 2.3-bit Compact build. The publisher's benchmark claims are not independently verified.",
+  "description": "Bad Theory Labs' agentic build of Ornith 1.0 35B, tuned for tool-heavy coding and long stretches of unattended work. Native reasoning, native tool calling, and a 256K context window that stays light on memory even when it's full. Capable-tier on-device option for machines with 32GB+ memory — see BTL-4 Compact for a smaller build.",
   "tags": [
     "badtheorylabs",
     "agentic",
@@ -16,8 +16,8 @@
     "name": "Bad Theory Labs",
     "url": "https://huggingface.co/badtheorylabs"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-08-06T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
@@ -136,7 +136,6 @@
   "llamaCpp": {
     "huggingfaceRepo": "bartowski/badtheorylabs_BTL-4-GGUF",
     "filename": "badtheorylabs_BTL-4-Q4_K_M.gguf",
-    "quantization": "Q4_K_M",
-    "residentBytes": 25669737830
+    "quantization": "Q4_K_M"
   }
 }

package/authoring/chat-models/btl4-compact-iq2.json

@@ -1,7 +1,7 @@
 {
   "id": "btl4-compact-iq2",
   "name": "BTL-4 Compact (35B-A3B, IQ2_XXS)",
-  "description": "Bad Theory Labs' agentic fine-tune of Ornith 1.0 35B, at 35B total parameters with ~2B active per token (Mixture-of-Experts: 256 experts, 8 routed). 30 of its 40 layers use linear attention, so the KV cache costs only ~20 KB/token and the full 256K context stays affordable on a laptop. This is a 2.3 bit-per-weight imatrix quantization: expert tensors at IQ2_XXS, everything else at Q4_K_M, which fits the whole model in under 10 GB. The publisher's benchmark claims are not independently verified.",
+  "description": "Bad Theory Labs' agentic build of Ornith 1.0 35B, tuned for tool-heavy coding and long stretches of unattended work. Native reasoning, native tool calling, and a 256K context window that stays light on memory even when it's full. Optimized for low-memory machines — the whole model fits in under 10GB, so it runs on a 16GB laptop, trading some accuracy against the full-size BTL-4 build.",
   "tags": [
     "badtheorylabs",
     "agentic",
@@ -16,8 +16,8 @@
     "name": "Bad Theory Labs",
     "url": "https://huggingface.co/badtheorylabs"
   },
-  "version": "1.0.1",
-  "updatedAt": "2026-08-06T00:00:00Z",
+  "version": "1.0.2",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
@@ -136,7 +136,6 @@
   "llamaCpp": {
     "huggingfaceRepo": "badtheorylabs/BTL-4-Compact",
     "filename": "BTL-4-IQ2_XXS.gguf",
-    "quantization": "IQ2_XXS",
-    "residentBytes": 11961559488
+    "quantization": "IQ2_XXS"
   }
 }

package/authoring/chat-models/deepseek-r1-8b-q4.json

@@ -12,8 +12,8 @@
     "name": "DeepSeek",
     "url": "https://ollama.com/library/deepseek-r1"
   },
-  "version": "1.1.0",
-  "updatedAt": "2026-04-26T00:00:00Z",
+  "version": "1.1.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "MIT",
   "parameterSize": "8B",
   "approxSizeBytes": 5200000000,

package/authoring/chat-models/gemma4-12b-q4.json

@@ -16,8 +16,8 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-12B-it"
   },
-  "version": "1.1.2",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.1.3",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "recoScore": 15,
   "parameterSize": "12B",
@@ -168,7 +168,6 @@
     "huggingfaceRepo": "unsloth/gemma-4-12B-it-qat-GGUF",
     "filename": "gemma-4-12B-it-qat-UD-Q4_K_XL.gguf",
     "quantization": "UD-Q4_K_XL",
-    "residentBytes": 8364078720,
     "draftModelFilename": "MTP/mtp-gemma-4-12B-it-Q4_0.gguf"
   },
   "releasePatch": {

package/authoring/chat-models/gemma4-12b-q8.json

@@ -15,8 +15,8 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-12B-it"
   },
-  "version": "1.0.2",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.0.3",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "recoScore": 15,
   "parameterSize": "12B",
@@ -142,7 +142,6 @@
     "huggingfaceRepo": "ggml-org/gemma-4-12B-it-GGUF",
     "filename": "gemma-4-12B-it-Q8_0.gguf",
     "quantization": "Q8_0",
-    "residentBytes": 15761707354,
     "draftModelFilename": "mtp-gemma-4-12B-it-Q8_0.gguf"
   },
   "releasePatch": {

package/authoring/chat-models/gemma4-26b-q4.json

@@ -16,8 +16,8 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-26B-A4B-it"
   },
-  "version": "1.2.1",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.2.2",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "recoScore": 15,
   "parameterSize": "25.2B",
@@ -68,7 +68,6 @@
     "huggingfaceRepo": "unsloth/gemma-4-26B-A4B-it-qat-GGUF",
     "filename": "gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf",
     "quantization": "UD-Q4_K_XL",
-    "residentBytes": 17401183718,
     "draftModelFilename": "MTP/mtp-gemma-4-26B-A4B-it-Q4_0.gguf"
   },
   "releasePatch": {

package/authoring/chat-models/gemma4-31b-q4.json

@@ -17,8 +17,8 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-31B-it"
   },
-  "version": "1.2.1",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.2.2",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "recoScore": 15,
   "parameterSize": "30.7B",
@@ -59,7 +59,6 @@
     "huggingfaceRepo": "unsloth/gemma-4-31B-it-qat-GGUF",
     "filename": "gemma-4-31B-it-qat-UD-Q4_K_XL.gguf",
     "quantization": "UD-Q4_K_XL",
-    "residentBytes": 21081151219,
     "draftModelFilename": "MTP/mtp-gemma-4-31B-it-Q4_0.gguf"
   },
   "releasePatch": {

package/authoring/chat-models/gemma4-e2b-q4.json

@@ -16,8 +16,8 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-E2B-it"
   },
-  "version": "1.2.0",
-  "updatedAt": "2026-07-31T00:00:00Z",
+  "version": "1.2.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "recoScore": 15,
   "parameterSize": "2.3B",
@@ -54,7 +54,6 @@
     "huggingfaceRepo": "unsloth/gemma-4-E2B-it-qat-GGUF",
     "filename": "gemma-4-E2B-it-qat-UD-Q4_K_XL.gguf",
     "quantization": "UD-Q4_K_XL",
-    "residentBytes": 3731251642,
     "draftModelFilename": "MTP/mtp-gemma-4-E2B-it-Q4_0.gguf"
   },
   "releasePatch": {

package/authoring/chat-models/gemma4-e4b-q4.json

@@ -15,8 +15,8 @@
     "name": "Google",
     "url": "https://huggingface.co/google/gemma-4-E4B-it"
   },
-  "version": "1.2.0",
-  "updatedAt": "2026-07-31T00:00:00Z",
+  "version": "1.2.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "recoScore": 15,
   "parameterSize": "4.5B",
@@ -47,7 +47,6 @@
     "huggingfaceRepo": "unsloth/gemma-4-E4B-it-qat-GGUF",
     "filename": "gemma-4-E4B-it-qat-UD-Q4_K_XL.gguf",
     "quantization": "UD-Q4_K_XL",
-    "residentBytes": 5940328250,
     "draftModelFilename": "MTP/mtp-gemma-4-E4B-it-Q4_0.gguf"
   },
   "releasePatch": {

package/authoring/chat-models/gpt-oss-20b-q4.json

@@ -14,8 +14,8 @@
     "name": "OpenAI",
     "url": "https://ollama.com/library/gpt-oss"
   },
-  "version": "1.1.0",
-  "updatedAt": "2026-04-26T00:00:00Z",
+  "version": "1.1.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "parameterSize": "20B",
   "approxSizeBytes": 12000000000,

package/authoring/chat-models/laguna-s-2.1-118b-q4.json

@@ -19,8 +19,8 @@
     "name": "Poolside",
     "url": "https://huggingface.co/poolside/Laguna-S-2.1"
   },
-  "version": "1.0.1",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.0.2",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "OpenMDW-1.1",
   "licenseClass": "open",
   "licenseShortName": "OpenMDW 1.1",
@@ -116,12 +116,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "poolside/Laguna-S-2.1-GGUF",
     "filename": "laguna-s-2.1-Q4_K_M.gguf",
-    "quantization": "Q4_K_M",
-    "residentBytes": 115238195712
+    "quantization": "Q4_K_M"
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/Laguna-S-2.1-oQ4e",
-    "quantization": "oQ4e",
-    "residentBytes": 70000000000
+    "quantization": "oQ4e"
   }
 }

package/authoring/chat-models/laguna-s-2.1-118b-q6.json

@@ -19,8 +19,8 @@
     "name": "Poolside",
     "url": "https://huggingface.co/poolside/Laguna-S-2.1"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-07-22T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "OpenMDW-1.1",
   "licenseClass": "open",
   "licenseShortName": "OpenMDW 1.1",
@@ -115,7 +115,6 @@
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/Laguna-S-2.1-oQ6e",
-    "quantization": "oQ6e",
-    "residentBytes": 99000000000
+    "quantization": "oQ6e"
   }
 }

package/authoring/chat-models/laguna-s-2.1-118b-q8.json

@@ -19,8 +19,8 @@
     "name": "Poolside",
     "url": "https://huggingface.co/poolside/Laguna-S-2.1"
   },
-  "version": "1.0.1",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.0.2",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "OpenMDW-1.1",
   "licenseClass": "open",
   "licenseShortName": "OpenMDW 1.1",
@@ -116,12 +116,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "poolside/Laguna-S-2.1-GGUF",
     "filename": "laguna-s-2.1-Q8_0.gguf",
-    "quantization": "Q8_0",
-    "residentBytes": 154500987802
+    "quantization": "Q8_0"
   },
   "mlx": {
     "huggingfaceRepo": "Vontra/Laguna-S-2.1-MLX-8bit",
-    "quantization": "8bit",
-    "residentBytes": 125000000000
+    "quantization": "8bit"
   }
 }

package/authoring/chat-models/lfm2.5-2.6b-q4.json

@@ -12,8 +12,8 @@
     "name": "Liquid AI",
     "url": "https://huggingface.co/LiquidAI/LFM2.5-2.6B"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-08-04T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "LFM Open License v1.0",
   "licenseClass": "custom-restricted",
   "licenseShortName": "LFM Open 1.0",

package/authoring/chat-models/llama3.2-3b-q4.json

@@ -11,8 +11,8 @@
     "name": "Meta",
     "url": "https://ollama.com/library/llama3.2"
   },
-  "version": "1.1.0",
-  "updatedAt": "2026-04-26T00:00:00Z",
+  "version": "1.1.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Llama 3.2 Community License",
   "parameterSize": "3B",
   "approxSizeBytes": 2019377696,

package/authoring/chat-models/mistral-7b-q4.json

@@ -11,8 +11,8 @@
     "name": "Mistral AI",
     "url": "https://ollama.com/library/mistral"
   },
-  "version": "1.1.0",
-  "updatedAt": "2026-04-26T00:00:00Z",
+  "version": "1.1.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "parameterSize": "7B",
   "approxSizeBytes": 4372812000,

package/authoring/chat-models/mistral-medium-3.5-128b-q4.json

@@ -14,8 +14,8 @@
     "name": "Mistral AI",
     "url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-05-22T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "MIT-Modified",
   "parameterSize": "128B",
   "approxSizeBytes": 72000000000,

package/authoring/chat-models/muse-glimmer-30b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "muse-glimmer-30b-q4",
   "name": "Muse Glimmer (30B, Q4)",
-  "description": "Meta Superintelligence Labs' Muse family, distilled from the larger Muse Spark down to 30B: a dense agentic model (29.6B language model plus a ~1.8B ViT-G/14 perception encoder) built for long-horizon tool use on consumer hardware. A new architecture rather than a Llama derivative — its own chat format and tool-call grammar. Multimodal (text + image), 131K native context, native tool calling, and a reasoning strength dial from low to xhigh. Meta's own calibrated K-quant build, sized to fit 24GB of VRAM.",
+  "description": "Meta Superintelligence Labs' Muse Glimmer at 30B parameters, distilled from the larger Muse Spark and built for long-horizon tool use on everyday hardware. Multimodal (text + image), 131K context, native tool calling, and adjustable reasoning effort. Meta's own build, sized for machines with 24GB+ memory.",
   "tags": [
     "meta",
     "agentic",
@@ -16,8 +16,8 @@
     "name": "Meta Superintelligence Labs",
     "url": "https://huggingface.co/meta-models/Muse-Glimmer-30B"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-08-10T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
@@ -120,7 +120,6 @@
     "huggingfaceRepo": "meta-models/Muse-Glimmer-30B-GGUF",
     "filename": "muse-glimmer-30B-kquant-17gb.gguf",
     "quantization": "K-Quant-17GB",
-    "mmprojFilename": "mmproj-kquant.gguf",
-    "residentBytes": 21416116224
+    "mmprojFilename": "mmproj-kquant.gguf"
   }
 }

package/authoring/chat-models/nemotron3-nano-30b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "nemotron3-nano-30b-q4",
   "name": "Nemotron-3 Nano (30B, MoE Reasoning)",
-  "description": "NVIDIA's Nemotron-3 Nano Omni — 30B-parameter hybrid Mamba/Transformer Mixture-of-Experts (~3B active per token) tuned for reasoning. Native function-calling, 128K context, multimodal (text + image + audio + video).",
+  "description": "NVIDIA's Nemotron-3 Nano Omni at 30B parameters, tuned for reasoning. Multimodal (text + image + audio + video), 128K context, native tool calling. Capable-tier on-device option for machines with 32GB+ memory.",
   "tags": [
     "nvidia",
     "reasoning",
@@ -18,8 +18,8 @@
     "name": "NVIDIA",
     "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-05-17T00:00:00Z",
+  "version": "1.0.2",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "NVIDIA Open Model License",
   "parameterSize": "30B",
   "approxSizeBytes": 24515129536,

package/authoring/chat-models/nemotron3-super-120b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "nemotron3-super-120b-q4",
   "name": "Nemotron-3 Super (120B-A12B)",
-  "description": "NVIDIA's flagship Nemotron-3 Super — 120B-parameter hybrid Mamba/Transformer Mixture-of-Experts (~12B active per token). Native function-calling, 128K context. Frontier-tier on-device reasoning for workstations and Mac Studio-class hardware — needs ~100GB of usable memory for Q4_K_M. MLX path is held until mlx-community publishes a build that doesn't require trust_remote_code; for now this is llama.cpp-only.",
+  "description": "NVIDIA's flagship Nemotron-3 Super at 120B parameters, tuned for reasoning. Native tool calling, 128K context. Frontier-tier on-device reasoning for workstations and Mac Studio-class hardware — needs ~100GB of usable memory. No Apple Silicon build yet.",
   "tags": [
     "nvidia",
     "reasoning",
@@ -15,8 +15,8 @@
     "name": "NVIDIA",
     "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-05-17T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "NVIDIA Open Model License",
   "parameterSize": "120B",
   "approxSizeBytes": 86051079584,

package/authoring/chat-models/nemotron3.5-lightning-30b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "nemotron3.5-lightning-30b-q4",
   "name": "Nemotron 3.5 Lightning (30B-A3B, Q4)",
-  "description": "NVIDIA's Nemotron 3.5 Lightning — a text-only 30B-parameter hybrid Mamba-2/Transformer Mixture-of-Experts model with about 3B active parameters per token, native reasoning and tool calling, and a 256K context window. The llama.cpp source is ggml-org's Q4_K_M build (about 25.4GB); Apple Silicon uses mlx-community's matching 4-bit affine conversion (about 17.8GB).",
+  "description": "NVIDIA's Nemotron 3.5 Lightning at 30B parameters — text-only, with native reasoning, native tool calling, and a 256K context window. Capable-tier on-device option: about 25GB of memory on most machines, about 18GB on Apple Silicon.",
   "tags": [
     "nvidia",
     "reasoning",
@@ -20,8 +20,8 @@
     "name": "NVIDIA",
     "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-08-11T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "minGezelVersion": "1.26223",
   "license": "OpenMDW-1.1",
   "licenseClass": "open",
@@ -124,12 +124,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "ggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
     "filename": "NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Q4_K_M.gguf",
-    "quantization": "Q4_K_M",
-    "residentBytes": 30516886733
+    "quantization": "Q4_K_M"
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-4bit",
-    "quantization": "4bit",
-    "residentBytes": 23130356200
+    "quantization": "4bit"
   }
 }

package/authoring/chat-models/nemotron3.5-lightning-30b-q8.json

@@ -1,7 +1,7 @@
 {
   "id": "nemotron3.5-lightning-30b-q8",
   "name": "Nemotron 3.5 Lightning (30B-A3B, Q8)",
-  "description": "NVIDIA's Nemotron 3.5 Lightning — a text-only 30B-parameter hybrid Mamba-2/Transformer Mixture-of-Experts model with about 3B active parameters per token, native reasoning and tool calling, and a 256K context window. This higher-fidelity variant uses ggml-org's Q8_0 llama.cpp build (about 35GB) or mlx-community's matching 8-bit affine conversion (about 33.6GB).",
+  "description": "NVIDIA's Nemotron 3.5 Lightning at 30B parameters — text-only, with native reasoning, native tool calling, and a 256K context window. 8-bit quantization for higher fidelity than the Q4 build, at roughly a third more memory (about 35GB, or 34GB on Apple Silicon).",
   "tags": [
     "nvidia",
     "reasoning",
@@ -20,8 +20,8 @@
     "name": "NVIDIA",
     "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-08-11T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "minGezelVersion": "1.26223",
   "license": "OpenMDW-1.1",
   "licenseClass": "open",
@@ -124,12 +124,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "ggml-org/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF",
     "filename": "NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Q8_0.gguf",
-    "quantization": "Q8_0",
-    "residentBytes": 42005570765
+    "quantization": "Q8_0"
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-8bit",
-    "quantization": "8bit",
-    "residentBytes": 43650255848
+    "quantization": "8bit"
   }
 }

package/authoring/chat-models/ornith-35b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "ornith-35b-q4",
   "name": "Ornith 1.0 (35B, Q4)",
-  "description": "DeepReinforce's Ornith 1.0 at 35B total parameters (Qwen 3.5 MoE architecture), an agentic-coding model trained with a self-improving reinforcement-learning framework — state-of-the-art on SWE-Bench and Terminal-Bench. Native <think> reasoning, OpenAI-style tool calling, 256K context. Q4_K_M quantization for a ~21GB footprint. MIT licensed.",
+  "description": "DeepReinforce's Ornith 1.0 at 35B parameters — an agentic-coding model built for hands-on software work: reading a codebase, running tools, and seeing a change through. Native reasoning, native tool calling, 256K context. Capable-tier on-device option for machines with 32GB+ memory. MIT licensed.",
   "tags": [
     "deepreinforce",
     "coding",
@@ -16,8 +16,8 @@
     "name": "DeepReinforce",
     "url": "https://huggingface.co/deepreinforce-ai"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-07-01T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "MIT",
   "licenseClass": "open",
   "licenseShortName": "MIT",
@@ -134,12 +134,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "deepreinforce-ai/Ornith-1.0-35B-GGUF",
     "filename": "ornith-1.0-35b-Q4_K_M.gguf",
-    "quantization": "Q4_K_M",
-    "residentBytes": 25400109312
+    "quantization": "Q4_K_M"
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/Ornith-1.0-35B-4bit",
-    "quantization": "4bit",
-    "residentBytes": 26557921365
+    "quantization": "4bit"
   }
 }

package/authoring/chat-models/ornith-35b-q8.json

@@ -1,7 +1,7 @@
 {
   "id": "ornith-35b-q8",
   "name": "Ornith 1.0 (35B, Q8)",
-  "description": "DeepReinforce's Ornith 1.0 at 35B total parameters (Qwen 3.5 MoE architecture), an agentic-coding model trained with a self-improving reinforcement-learning framework — state-of-the-art on SWE-Bench and Terminal-Bench. Native <think> reasoning, OpenAI-style tool calling, 256K context. Q8_0 quantization for higher fidelity than the Q4 build, at roughly double the memory footprint (~37GB). MIT licensed.",
+  "description": "DeepReinforce's Ornith 1.0 at 35B parameters — an agentic-coding model built for hands-on software work: reading a codebase, running tools, and seeing a change through. Native reasoning, native tool calling, 256K context. 8-bit quantization for higher fidelity than the Q4 build, at roughly double the memory footprint (~37GB). MIT licensed.",
   "tags": [
     "deepreinforce",
     "coding",
@@ -16,8 +16,8 @@
     "name": "DeepReinforce",
     "url": "https://huggingface.co/deepreinforce-ai"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-07-01T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "MIT",
   "licenseClass": "open",
   "licenseShortName": "MIT",
@@ -134,12 +134,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "deepreinforce-ai/Ornith-1.0-35B-GGUF",
     "filename": "ornith-1.0-35b-Q8_0.gguf",
-    "quantization": "Q8_0",
-    "residentBytes": 44283766656
+    "quantization": "Q8_0"
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/Ornith-1.0-35B-8bit",
-    "quantization": "8bit",
-    "residentBytes": 49072879453
+    "quantization": "8bit"
   }
 }

package/authoring/chat-models/ornith-9b-q4.json

@@ -1,7 +1,7 @@
 {
   "id": "ornith-9b-q4",
   "name": "Ornith 1.0 (9B)",
-  "description": "DeepReinforce's Ornith 1.0 at 9B parameters (dense, built on Qwen 3.5 + Gemma 4), an agentic-coding model trained with a self-improving reinforcement-learning framework. Native <think> reasoning, OpenAI-style tool calling, 256K context. The small-tier pick for tool-heavy work on a 12GB+ inference budget. Q4_K_M quantization. MIT licensed.",
+  "description": "DeepReinforce's Ornith 1.0 at 9B parameters — an agentic-coding model built for hands-on software work. Native reasoning, native tool calling, 256K context. The small-tier pick for tool-heavy work on a 12GB+ inference budget. MIT licensed.",
   "tags": [
     "deepreinforce",
     "coding",
@@ -15,8 +15,8 @@
     "name": "DeepReinforce",
     "url": "https://huggingface.co/deepreinforce-ai"
   },
-  "version": "1.0.0",
-  "updatedAt": "2026-07-01T00:00:00Z",
+  "version": "1.0.1",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "MIT",
   "licenseClass": "open",
   "licenseShortName": "MIT",
@@ -135,12 +135,10 @@
   "llamaCpp": {
     "huggingfaceRepo": "deepreinforce-ai/Ornith-1.0-9B-GGUF",
     "filename": "ornith-1.0-9b-Q4_K_M.gguf",
-    "quantization": "Q4_K_M",
-    "residentBytes": 6754930445
+    "quantization": "Q4_K_M"
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/Ornith-1.0-9B-4bit",
-    "quantization": "4bit",
-    "residentBytes": 7770193739
+    "quantization": "4bit"
   }
 }

package/authoring/chat-models/qwen3.5-122b-a10b-q4.json

@@ -18,8 +18,8 @@
     "name": "Alibaba",
     "url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B"
   },
-  "version": "1.0.1",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.0.3",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "licenseClass": "open",
   "licenseShortName": "Apache 2.0",
@@ -149,7 +149,6 @@
     "huggingfaceRepo": "unsloth/Qwen3.5-122B-A10B-MTP-GGUF",
     "shardsDir": "UD-Q4_K_M",
     "quantization": "UD-Q4_K_M",
-    "residentBytes": 93912483840,
     "mmprojFilename": "mmproj-BF16.gguf"
   },
   "releasePatch": {
@@ -166,7 +165,6 @@
   },
   "mlx": {
     "huggingfaceRepo": "mlx-community/Qwen3.5-122B-A10B-4bit",
-    "quantization": "4bit",
-    "residentBytes": 90506802715
+    "quantization": "4bit"
   }
 }

package/authoring/chat-models/qwen3.5-2b-q4.json

@@ -14,8 +14,8 @@
     "name": "Alibaba",
     "url": "https://ollama.com/library/qwen3.5"
   },
-  "version": "1.1.2",
-  "updatedAt": "2026-07-28T00:00:00Z",
+  "version": "1.1.3",
+  "updatedAt": "2026-08-15T00:00:00Z",
   "license": "Apache-2.0",
   "parameterSize": "2B",
   "approxSizeBytes": 1329851808,
@@ -48,8 +48,7 @@
   "llamaCpp": {
     "huggingfaceRepo": "unsloth/Qwen3.5-2B-MTP-GGUF",
     "filename": "Qwen3.5-2B-Q4_K_M.gguf",
-    "quantization": "Q4_K_M",
-    "residentBytes": 1595822170
+    "quantization": "Q4_K_M"
   },
   "releasePatch": {
     "tuning": {