chore: Narrow jina-api known models to v2-base-code and code-embeddings family

Drops jina-embeddings-v3, v2-base-en, and clip-v2 from the known-models list
and adds verified native dimensions (and Matryoshka sizes in the comment) for
jina-code-embeddings-0.5b (896) and jina-code-embeddings-1.5b (1536). The
task-aware prefix is now scoped to the code-embeddings family.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: GoodbyePlanet

README.md

@@ -246,8 +246,8 @@ configured model — no need to set dimensions manually unless you want to overr
 | `JINA_MODEL`             | `jinaai/jina-embeddings-v2-base-code` | `jina`     | Informational only — the TEI container's `--model-id` flag is what actually loads. Edit `docker-compose.yaml` to change models. |
 | `JINA_DIMENSIONS`        | `768`                                 | `jina`     | Vector dimensions of the TEI model                                                                                              |
 | `JINA_API_KEY`           | *(required if provider=jina-api)*     | `jina-api` | Jina AI API key (hosted endpoint at `api.jina.ai`)                                                                              |
-| `JINA_API_MODEL`         | `jina-embeddings-v2-base-code`        | `jina-api` | Hosted Jina model — also supports `jina-embeddings-v3`, `jina-code-embeddings-0.5b`, `jina-code-embeddings-1.5b`                 |
-| `JINA_API_DIMENSIONS`    | *(native)*                            | `jina-api` | Optional Matryoshka override (v3 and code-embeddings models support shrinking); required for models without a native default    |
+| `JINA_API_MODEL`         | `jina-embeddings-v2-base-code`        | `jina-api` | Hosted Jina model — also supports `jina-code-embeddings-0.5b`, `jina-code-embeddings-1.5b`                                      |
+| `JINA_API_DIMENSIONS`    | *(native)*                            | `jina-api` | Optional Matryoshka override (code-embeddings models support shrinking); required for models without a native default          |
 | `VOYAGE_API_KEY`         | *(required if provider=voyage)*       | `voyage`   | Voyage AI API key                                                                                                               |
 | `VOYAGE_MODEL`           | `voyage-code-3`                       | `voyage`   | Voyage embedding model                                                                                                          |
 | `VOYAGE_DIMENSIONS`      | *(native)*                            | `voyage`   | Optional override — Voyage code-3 supports `256` / `512` / `1024` / `2048`                                                      |

server/embeddings/jina_api.py

@@ -16,24 +16,24 @@
 _BATCH_SIZE = 128
 _BACKOFF_DELAYS = [10, 20, 30, 40]
 
-# Native output dimensions for known models. v3 and the jina-code-embeddings
-# family support Matryoshka truncation via the `dimensions` API parameter —
-# override with JINA_API_DIMENSIONS.
-#
-# Other code-tuned models available on the hosted API:
-#   - jina-code-embeddings-0.5b
-#   - jina-code-embeddings-1.5b
-# Their native dimensions vary; set JINA_API_DIMENSIONS to declare the size.
+# Native output dimensions for known models. The jina-code-embeddings family
+# supports Matryoshka truncation via the `dimensions` API parameter —
+# override with JINA_API_DIMENSIONS to one of the supported sizes:
+#   - jina-code-embeddings-0.5b: 64, 128, 256, 512, 896 (native)
+#     https://jina.ai/models/jina-code-embeddings-0.5b
+#   - jina-code-embeddings-1.5b: 128, 256, 512, 1024, 1536 (native)
+#     https://jina.ai/models/jina-code-embeddings-1.5b
+# jina-embeddings-v2-base-code is fixed-size and does not support truncation.
+#     https://jina.ai/models/jina-embeddings-v2-base-code
 _NATIVE_DIMENSIONS: dict[str, int] = {
     "jina-embeddings-v2-base-code": 768,
-    "jina-embeddings-v2-base-en": 768,
-    "jina-embeddings-v3": 1024,
-    "jina-clip-v2": 1024,
+    "jina-code-embeddings-0.5b": 896,
+    "jina-code-embeddings-1.5b": 1536,
 }
 
-# Models that accept the `task` parameter (asymmetric retrieval). v2 models
-# are single-mode and reject `task`, so we omit it for them.
-_TASK_AWARE_PREFIXES = ("jina-embeddings-v3", "jina-code-embeddings-")
+# Models that accept the `task` parameter (asymmetric retrieval). The v2 model
+# is single-mode and rejects `task`, so we omit it.
+_TASK_AWARE_PREFIXES = ("jina-code-embeddings-",)
 
 
 class JinaApiEmbeddingProvider(EmbeddingProvider):

tests/embeddings/test_jina_api.py

@@ -37,7 +37,7 @@ def test_missing_api_key_raises(monkeypatch):
 
 def test_unknown_model_without_dimensions_raises(monkeypatch):
     monkeypatch.setattr(settings, "jina_api_key", "k")
-    monkeypatch.setattr(settings, "jina_api_model", "jina-code-embeddings-0.5b")
+    monkeypatch.setattr(settings, "jina_api_model", "jina-future-99")
     monkeypatch.setattr(settings, "jina_api_dimensions", None)
     with pytest.raises(RuntimeError, match="JINA_API_DIMENSIONS"):
         JinaApiEmbeddingProvider()
@@ -50,7 +50,7 @@ def test_dimensions_native(jina_api_settings):
 
 def test_dimensions_override(monkeypatch):
     monkeypatch.setattr(settings, "jina_api_key", "k")
-    monkeypatch.setattr(settings, "jina_api_model", "jina-embeddings-v3")
+    monkeypatch.setattr(settings, "jina_api_model", "jina-code-embeddings-1.5b")
     monkeypatch.setattr(settings, "jina_api_dimensions", 512)
     p = JinaApiEmbeddingProvider()
     assert p.dimensions == 512
@@ -81,12 +81,12 @@ async def test_embed_query_omits_task_for_v2_model(provider):
 
 
 @respx.mock
-async def test_embed_batch_sends_task_for_v3_model(monkeypatch):
+async def test_embed_batch_sends_task_for_code_embeddings_model(monkeypatch):
     monkeypatch.setattr(settings, "jina_api_key", "test-key")
-    monkeypatch.setattr(settings, "jina_api_model", "jina-embeddings-v3")
+    monkeypatch.setattr(settings, "jina_api_model", "jina-code-embeddings-1.5b")
     monkeypatch.setattr(settings, "jina_api_dimensions", None)
     route = respx.post("https://api.jina.ai/v1/embeddings").mock(
-        return_value=httpx.Response(200, json=_vectors_response(["a"], dim=1024))
+        return_value=httpx.Response(200, json=_vectors_response(["a"], dim=1536))
     )
     p = JinaApiEmbeddingProvider()
     try:
@@ -98,12 +98,12 @@ async def test_embed_batch_sends_task_for_v3_model(monkeypatch):
 
 
 @respx.mock
-async def test_embed_query_sends_task_for_v3_model(monkeypatch):
+async def test_embed_query_sends_task_for_code_embeddings_model(monkeypatch):
     monkeypatch.setattr(settings, "jina_api_key", "test-key")
-    monkeypatch.setattr(settings, "jina_api_model", "jina-embeddings-v3")
+    monkeypatch.setattr(settings, "jina_api_model", "jina-code-embeddings-1.5b")
     monkeypatch.setattr(settings, "jina_api_dimensions", None)
     route = respx.post("https://api.jina.ai/v1/embeddings").mock(
-        return_value=httpx.Response(200, json=_vectors_response(["q"], dim=1024))
+        return_value=httpx.Response(200, json=_vectors_response(["q"], dim=1536))
     )
     p = JinaApiEmbeddingProvider()
     try:
@@ -114,23 +114,6 @@ async def test_embed_query_sends_task_for_v3_model(monkeypatch):
     assert body["task"] == "retrieval.query"
 
 
-@respx.mock
-async def test_embed_batch_sends_task_for_code_embeddings_model(monkeypatch):
-    monkeypatch.setattr(settings, "jina_api_key", "test-key")
-    monkeypatch.setattr(settings, "jina_api_model", "jina-code-embeddings-1.5b")
-    monkeypatch.setattr(settings, "jina_api_dimensions", 1024)
-    route = respx.post("https://api.jina.ai/v1/embeddings").mock(
-        return_value=httpx.Response(200, json=_vectors_response(["a"], dim=1024))
-    )
-    p = JinaApiEmbeddingProvider()
-    try:
-        await p.embed_batch(["a"])
-    finally:
-        await p.close()
-    body = json.loads(route.calls.last.request.read())
-    assert body["task"] == "retrieval.passage"
-
-
 @respx.mock
 async def test_embed_batch_chunks_at_128(provider):
     route = respx.post("https://api.jina.ai/v1/embeddings").mock(
@@ -159,7 +142,7 @@ async def test_authorization_header_set(provider):
 @respx.mock
 async def test_embed_batch_passes_dimensions_when_overridden(monkeypatch):
     monkeypatch.setattr(settings, "jina_api_key", "test-key")
-    monkeypatch.setattr(settings, "jina_api_model", "jina-embeddings-v3")
+    monkeypatch.setattr(settings, "jina_api_model", "jina-code-embeddings-1.5b")
     monkeypatch.setattr(settings, "jina_api_dimensions", 256)
 
     route = respx.post("https://api.jina.ai/v1/embeddings").mock(