diff --git a/.github/ci/benchmark_config.json b/.github/ci/benchmark_config.json
index 43f8895d..0700669b 100644
--- a/.github/ci/benchmark_config.json
+++ b/.github/ci/benchmark_config.json
@@ -315,6 +315,9 @@
     },
     "smolvlm_500m": {
       "config": "ported_models/llama_cpp_et/benchmarks/smolvlm_500m.json"
+    },
+    "modernbert_sentiment": {
+      "config": "ported_models/llama_cpp_et/benchmarks/modernbert_sentiment.json"
     }
   }
 }
diff --git a/ported_models/llama_cpp_et/artifacts.json b/ported_models/llama_cpp_et/artifacts.json
index 9c419ae1..614e6a55 100644
--- a/ported_models/llama_cpp_et/artifacts.json
+++ b/ported_models/llama_cpp_et/artifacts.json
@@ -485,6 +485,25 @@
       "sha256": "d1eb8b6b23979205fdf63703ed10f788131a3f812c7b1f72e0119d5d81295150",
       "size_bytes": 108783360,
       "note": "SmolVLM 500M vision projector (SigLIP ~93M + MLP). Q8_0 quantized. Must be loaded alongside smolvlm_500m_q8_gguf."
+    },
+    "modernbert_sentiment_f32_gguf": {
+      "kind": "model",
+      "framework": "llama.cpp-et",
+      "variant": "modernBERT-base-multilingual-sentiment-F32",
+      "filename": "modernbert_sentiment_f32.gguf",
+      "env": "MODERNBERT_SENTIMENT_MODEL_PATH",
+      "source": {
+        "type": "github_release",
+        "repo": "DarthCeltic/hf-hackathon",
+        "tag": "model-port-modernbert-sentiment-v1",
+        "filename": "modernbert_sentiment_f32.gguf",
+        "url": "https://github.com/DarthCeltic/hf-hackathon/releases/download/model-port-modernbert-sentiment-v1/modernbert_sentiment_f32.gguf",
+        "note": "GGUF conversion of upstream clapAI/modernBERT-base-multilingual-sentiment (huggingface.co, revision baa7f26fee4eca5aec35538caa5fdda7b86d83c6, license apache-2.0), produced with this repo's own convert_hf_to_gguf.py (ggml LLM_ARCH_MODERN_BERT). Hosted as a GitHub Release asset because we do not have a Hugging Face upload token; reproducible from the committed recipe (ported_models/modernbert_sentiment/docs/RECIPE.md)."
+      },
+      "sha256": "7896d82f2548762409157ae4e68732cd2c78796b5c8cea5e0145b964e916ab4c",
+      "size": 600207808,
+      "local_cache": "local-artifacts/models/modernbert_sentiment_f32.gguf",
+      "board_path": "/data/models/modernbert_sentiment_f32.gguf"
     }
   }
 }
diff --git a/ported_models/llama_cpp_et/benchmarks/modernbert_sentiment.json b/ported_models/llama_cpp_et/benchmarks/modernbert_sentiment.json
new file mode 100644
index 00000000..1d341432
--- /dev/null
+++ b/ported_models/llama_cpp_et/benchmarks/modernbert_sentiment.json
@@ -0,0 +1,34 @@
+{
+  "runner": "llama_server",
+  "board": true,
+  "framework": {
+    "name": "llama.cpp-et",
+    "runner": "llama_server",
+    "source_artifact": "llama_cpp_source"
+  },
+  "artifacts_file": "../artifacts.json",
+  "canonical_variant": "modernbert_sentiment_f32",
+  "score": {
+    "metric": "tokens_per_second",
+    "label": "Decode tokens/s",
+    "higher_is_better": true
+  },
+  "llama_server": {
+    "source_artifact": "llama_cpp_source",
+    "model_artifact": "modernbert_sentiment_f32_gguf",
+    "server_artifact": "llama_server",
+    "workdir_artifact": "llama_cpp_build",
+    "host": "127.0.0.1",
+    "port": 18152,
+    "device": "ET",
+    "gpu_layers": 99,
+    "ctx_size": 512,
+    "batch_size": 128,
+    "ubatch_size": 128,
+    "parallel": 1,
+    "cache_ram_mib": 0,
+    "ready_timeout_s": 300,
+    "request_timeout_s": 420,
+    "api": "rerank"
+  }
+}
diff --git a/ported_models/modernbert_sentiment/docs/RECIPE.md b/ported_models/modernbert_sentiment/docs/RECIPE.md
new file mode 100644
index 00000000..8892d8d4
--- /dev/null
+++ b/ported_models/modernbert_sentiment/docs/RECIPE.md
@@ -0,0 +1,78 @@
+# ModernBERT-base Multilingual Sentiment Porting Recipe
+
+## Overview
+
+Adds `clapAI/modernBERT-base-multilingual-sentiment` (149M-parameter
+encoder-only sequence-classification model, 3-way sentiment) to the
+`llama.cpp-et` framework. Confirmed via local GGUF metadata inspection and
+a real server load test: `general.architecture = modern-bert` -- distinct
+from `bert` (already ported earlier in this campaign as `distilbert_sst2`)
+and from `nomic-bert`/`jina-bert-v2`/`jina-bert-v3`/`neo-bert`/`eurobert`
+(all separate, unclaimed arches) -- ModernBERT is architecturally its own
+family: alternating global/local sliding-window attention (`n_swa=128`,
+pattern every 3 layers), RoPE instead of learned position embeddings,
+GeGLU FFN.
+
+This artifact was self-converted earlier this session (before this
+specific porting campaign) and hosted as a GitHub Release
+(`model-port-modernbert-sentiment-v1`), matching the same precedent as
+`distilbert_sst2`/`roberta_sst2` -- it had not yet had a model-ports claim
+filed until now.
+
+## Model Reference
+
+- **Source**: `clapAI/modernBERT-base-multilingual-sentiment` (Hugging
+  Face), revision `baa7f26fee4eca5aec35538caa5fdda7b86d83c6`
+- **License**: Apache-2.0
+- **Artifact**: self-converted F32 GGUF (no quantization -- encoder
+  classification heads are small; F32 avoids any quantization-accuracy
+  question for a 3-way classifier), produced via this repo's own
+  `convert_hf_to_gguf.py`, hosted at
+  `https://github.com/DarthCeltic/hf-hackathon/releases/download/model-port-modernbert-sentiment-v1/modernbert_sentiment_f32.gguf`,
+  `sha256=7896d82f2548762409157ae4e68732cd2c78796b5c8cea5e0145b964e916ab4c`
+  (verified locally against the downloaded file), 600,207,808 bytes
+- **Architecture**: `arch = modern-bert` per GGUF metadata, 138 tensors,
+  22 layers, 768 embedding, 12 heads, sliding-window attention (window
+  128, every 3rd layer is global), `n_cls_out=3`
+  (`cls_label = [negative, neutral, positive]`)
+
+## Verification performed this round
+
+Built `llama-server` from a plain CPU-only (`GGML_ET=OFF`) configuration
+of the vendored `llama.cpp-et` source, loaded the GGUF, and issued a real
+`POST /rerank` request. The server loaded all metadata correctly (3-way
+classifier head, sliding-window pattern) and returned real relevance
+scores for both documents -- confirmed live, not just GGUF-metadata
+inspection:
+
+```
+{"model":"modernbert_sentiment_f32.gguf","results":[
+  {"index":1,"relevance_score":2.3337833881378174},
+  {"index":0,"relevance_score":0.5375099182128906}]}
+```
+
+This is the same verification tier as `distilbert_sst2`/`roberta_sst2`
+earlier in this campaign (real server load + real endpoint response), not
+just a metadata read.
+
+## Why this port likely needs no new ET-SoC1 kernel work
+
+Sliding-window attention is already proven on this backend by Gemma2/
+Cohere2/Starcoder2/PLaMo-3 earlier in this campaign (an ordinary masked
+`GGML_OP_SOFT_MAX`, not a new op). RoPE, LayerNorm-family norm, and GeGLU
+FFN (an existing gated-unary op) are all already proven elsewhere on this
+board. Not confirmed live against ET sysemu specifically this round --
+only the CPU backend was exercised.
+
+## Open items for maintainer review
+
+- Registered in `artifacts.json`, `ported_models/llama_cpp_et/benchmarks/modernbert_sentiment.json`,
+  and `.github/ci/benchmark_config.json` -- but see the same caveat as
+  `distilbert_sst2`: the shared runner (`run_llama_server_benchmark.py`)
+  only speaks `/completion`/`/v1/chat/completions`, not `/rerank`, so this
+  entry documents the port but cannot be scored by automated per-PR CI
+  without a maintainer adding `/rerank` support to that script (a protected
+  file we cannot edit ourselves).
+- `ported_models/submissions/model_ports/modernbert_sentiment.json` is the
+  model-ports track claim, pending identity approval.
+- No changes to any protected file or the vendored submodule.
diff --git a/ported_models/modernbert_sentiment/oracle/sentiment_oracle.json b/ported_models/modernbert_sentiment/oracle/sentiment_oracle.json
new file mode 100644
index 00000000..debbae37
--- /dev/null
+++ b/ported_models/modernbert_sentiment/oracle/sentiment_oracle.json
@@ -0,0 +1,27 @@
+{
+  "oracle_type": "sentiment_rerank_reference",
+  "model_artifact": {
+    "repo": "DarthCeltic/hf-hackathon",
+    "release_tag": "model-port-modernbert-sentiment-v1",
+    "filename": "modernbert_sentiment_f32.gguf",
+    "sha256": "7896d82f2548762409157ae4e68732cd2c78796b5c8cea5e0145b964e916ab4c"
+  },
+  "reference_run": {
+    "endpoint": "POST /rerank",
+    "results": [
+      {
+        "index": 1,
+        "relevance_score": 2.3337833881378174
+      },
+      {
+        "index": 0,
+        "relevance_score": 0.5375099182128906
+      }
+    ],
+    "measured_on": "CPU (ggml-cpu backend, GGML_ET=OFF build)"
+  },
+  "comparison_threshold": {
+    "metric": "relevance_score_rank_order",
+    "note": "A reproduction run should preserve the same relative rank order (index 1 scoring higher than index 0) for the same two fixed input documents; exact score reproduction is not expected across backends but rank order should be stable. No ET-SoC1 hardware was available to this session to perform a full-offload re-run directly."
+  }
+}
diff --git a/ported_models/submissions/model_ports/modernbert_sentiment.json b/ported_models/submissions/model_ports/modernbert_sentiment.json
new file mode 100644
index 00000000..ad1b7948
--- /dev/null
+++ b/ported_models/submissions/model_ports/modernbert_sentiment.json
@@ -0,0 +1,14 @@
+{
+  "schema_version": 1,
+  "track": "most_models_ported",
+  "benchmark_model": "modernbert_sentiment",
+  "identity_id": "modern-bert",
+  "source": {
+    "repo": "clapAI/modernBERT-base-multilingual-sentiment",
+    "revision": "baa7f26fee4eca5aec35538caa5fdda7b86d83c6",
+    "license": "apache-2.0"
+  },
+  "implementation_paths": ["ported_models/modernbert_sentiment"],
+  "benchmark_config": ".github/ci/benchmark_config.json",
+  "recipe": "ported_models/modernbert_sentiment/docs/RECIPE.md"
+}