Add Mixedbread AI Rerank support #140477

Evgenii-Kazannik · 2026-01-09T21:37:13Z

RERANK

request put {{base-url}}/_inference/rerank/mixedbread

{ "service": "mixedbread", "service_settings": { "api_key": "{{mb-api-key}}", "model_id": "mixedbread-ai/mxbai-rerank-xsmall-v1" }, "task_settings": { "return_documents": true, "top_k": 1 } }

response

{ "inference_id": "mixedbread", "task_type": "rerank", "service": "mixedbread", "service_settings": { "model_id": "mixedbread-ai/mxbai-rerank-xsmall-v1", "rate_limit": { "requests_per_minute": 240 } }, "task_settings": { "return_documents": true } }

request post {{base-url}}/_inference/rerank/mixedbread

{ "input": ["Luke", "like", "leia", "chewy","r2d2", "star", "wars"], "query": "star wars main character", "top_n": 2, "return_documents": true }

response

{ "rerank": [ { "index": 0, "relevance_score": 0.083740234, "text": "Luke" }, { "index": 2, "relevance_score": 0.06994629, "text": "leia" } ] }

direct request post https://api.mixedbread.com/v1/reranking

{ "model": "mixedbread-ai/mxbai-rerank-xsmall-v1", "query": "Who is the author of To Kill a Mockingbird?", "input": [ "To Kill a Mockingbird is a novel by Harper Lee", "The novel Moby-Dick was written by Herman Melville", "Harper Lee, an American novelist", "Jane Austen was an English novelist", "The Harry Potter series written by British author J.K. Rowling", "The Great Gatsby, a novel written by American author F. Scott Fitzgerald" ], "top_k": 3, "return_input": true }

response

"usage": {
    "prompt_tokens": 162,
    "total_tokens": 162,
    "completion_tokens": 0
},
"model": "mixedbread-ai/mxbai-rerank-xsmall-v1",
"data": [
    {
        "index": 0,
        "score": 0.98291015625,
        "input": "To Kill a Mockingbird is a novel by Harper Lee",
        "object": "rank_result"
    },
    {
        "index": 2,
        "score": 0.61962890625,
        "input": "Harper Lee, an American novelist",
        "object": "rank_result"
    },
    {
        "index": 3,
        "score": 0.36328125,
        "input": "Jane Austen was an English novelist",
        "object": "rank_result"
    }
],
"object": "list",
"top_k": 3,
"return_input": true

}

elasticsearchmachine · 2026-01-20T22:19:20Z

Pinging @elastic/search-inference-team (Team:Search - Inference)

Copilot

Pull request overview

This pull request adds support for Mixedbread AI's rerank API to Elasticsearch's inference plugin. The implementation follows the established pattern for inference service providers and includes comprehensive test coverage.

Changes:

Implements Mixedbread rerank service with model, request/response handling, and action creators
Adds service settings and task settings with configurable parameters (top_n, return_documents)
Registers the new service in InferencePlugin and InferenceNamedWriteablesProvider

Reviewed changes

Copilot reviewed 27 out of 27 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
MixedbreadService.java	Main service implementation for Mixedbread rerank with configuration and inference methods
MixedbreadRerankModel.java	Model class defining rerank-specific configuration and URI building
MixedbreadRerankRequest.java	Request builder for Mixedbread rerank API calls
MixedbreadRerankResponseEntity.java	Response parser for Mixedbread rerank API responses
MixedbreadRerankTaskSettings.java	Task-level settings (top_n, return_documents)
MixedbreadRerankServiceSettings.java	Service-level settings (model_id, rate limits)
MixedbreadActionCreator.java	Creates executable actions for rerank operations
MixedbreadConstants.java	Shared constants for field names and API paths
MixedbreadAccount.java	Account credentials and URI management
InferencePlugin.java	Registers the Mixedbread service factory
InferenceNamedWriteablesProvider.java	Registers named writeables for serialization
Test files (8 files)	Comprehensive test coverage for all components

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-01-21T13:51:32Z

...sticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankTaskSettingsTests.java

+        assertThat(thrownException.getMessage(), containsString("field [top_n] is not of the expected type"));
+    }
+
+    public void UpdatedTaskSettings_WithEmptyMap_ReturnsSameSettings() {


The test name has inconsistent capitalization. It should start with 'test' (lowercase) to match Java naming conventions and be consistent with other test methods in the same file.

Suggested change

public void UpdatedTaskSettings_WithEmptyMap_ReturnsSameSettings() {

public void testUpdatedTaskSettings_WithEmptyMap_ReturnsSameSettings() {

Copilot · 2026-01-21T13:51:32Z

...org/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankModelTests.java

+
+import java.util.Map;
+
+import static org.elasticsearch.xpack.inference.services.jinaai.rerank.JinaAIRerankTaskSettingsTests.getTaskSettingsMap;


The test is incorrectly using a JinaAI import for the helper method. This should use the Mixedbread equivalent method 'MixedbreadRerankTaskSettingsTests.getTaskSettingsMap' instead of importing from JinaAI rerank task settings tests.

Suggested change

import static org.elasticsearch.xpack.inference.services.jinaai.rerank.JinaAIRerankTaskSettingsTests.getTaskSettingsMap;

import static org.elasticsearch.xpack.inference.services.mixedbread.rerank.MixedbreadRerankTaskSettingsTests.getTaskSettingsMap;

Copilot · 2026-01-21T13:51:33Z

...asticsearch/xpack/inference/services/mixedbread/response/MixedbreadRerankResponseEntity.java

+        XContentParser.Token token = parser.currentToken();
+        ensureExpectedToken(XContentParser.Token.START_OBJECT, token, parser);
+
+        positionParserAtTokenAfterField(parser, "data", "FAILED_TO_FIND_FIELD_TEMPLATE"); // TODO error message


There's a TODO comment with an incomplete error message placeholder. The error message template should be properly defined, as this is used when parsing fails if the 'data' field is not found in the response.

Suggested change

positionParserAtTokenAfterField(parser, "data", "FAILED_TO_FIND_FIELD_TEMPLATE"); // TODO error message

positionParserAtTokenAfterField(parser, "data", "Failed to find [data] field in Mixedbread rerank response");

Copilot · 2026-01-21T13:51:33Z

...g/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankTaskSettings.java

+    public Boolean getDoesReturnDocuments() {
+        return returnDocuments;
+    }
+
+    public Integer getTopNDocumentsOnly() {
+        return topNDocumentsOnly;
+    }
+
+    public Boolean getReturnDocuments() {
+        return returnDocuments;
+    }


The class has two redundant getter methods for the same field. Both 'getDoesReturnDocuments()' and 'getReturnDocuments()' return the same 'returnDocuments' field. One of these methods should be removed to avoid confusion and maintain clean API design.

I agree, let's remove one of these.

Copilot · 2026-01-21T13:51:33Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+
+    @Override
+    public int rerankerWindowSize(String modelId) {
+        // Cohere rerank model truncates at 4096 tokens https://docs.cohere.com/reference/rerank


The comment incorrectly mentions "Cohere rerank model" when this is a Mixedbread service implementation. The comment should be updated to reference Mixedbread's actual model token limits or window size documentation.

Suggested change

// Cohere rerank model truncates at 4096 tokens https://docs.cohere.com/reference/rerank

// Mixedbread rerank models currently support context windows of up to 4096 tokens (see Mixedbread documentation)

This is pointing to cohere, I think we want the numbers posted here: https://www.mixedbread.com/docs/models/reranking/mxbai-rerank-large-v2

Looks like the older models have a window size of 512. We should make this configurable though. Let's add an optional field to the service settings that can control this value and default it to 8k.

The newly added settings field will need to be used in

@Override public int rerankerWindowSize(String modelId) {

so we probably need to pass a model as a parameter instead of modelId but then I need to make refactoring impacting other services, namely to make the change in TransportGetRerankerWindowSizeAction and services overriding rerankerWindowSize

Should we do it or it's better to revert some changes and make it configurable via model_id like that?

MixedbreadService

```

private static final Map<String, Integer> RERANKERS_INPUT_SIZE = Map.of(
"mixedbread-ai/mxbai-rerank-xsmall-v1",
512,
"mixedbread-ai/mxbai-rerank-base-v1",
512,
"mixedbread-ai/mxbai-rerank-large-v1",
512
// Windows size.
// The v1 models: 512
// The v2 models: at least 8k
// https://www.mixedbread.com/docs/models/reranking/mxbai-rerank-large-v1
);

@Override public int rerankerWindowSize(String modelId) { Integer inputSize = RERANKERS_INPUT_SIZE.get(modelId); return inputSize != null ? inputSize : DEFAULT_RERANKER_INPUT_SIZE_WORDS; }

@jonathan-buttner @DonalEvans

The window size is not something a user can configure, it's an unchanging property of the model, so we don't need to store it in service settings. The current approach of having a map with the model IDs that don't use the 8k default is fine, but the rerankerWindowSize() method returns the size in words, not in tokens, so we'll need to translate from the 512/8000 values in tokens to smaller values in words by multiplying by 0.75 and rounding down a bit, which is the approach we use for other providers.

For consistency, the PR that originally introduced this feature can be used as a guide, with 512 tokens translating to a window size of 300 words, and 8000 tokens translating to 5500 words.

Copilot · 2026-01-21T13:51:33Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+        TimeValue timeout,
+        ActionListener<List<ChunkedInference>> listener
+    ) {
+


The method 'doChunkedInfer' has an empty implementation. This should either throw an UnsupportedOperationException with a descriptive message, or contain a proper implementation if chunked inference is supported.

Suggested change

throw new UnsupportedOperationException("Chunked inference is not supported for service [" + NAME + "]");

Agreed, let's throw something like:

throw new UnsupportedOperationException(Strings.format("%s service does not support chunked inference", NAME));

Copilot · 2026-01-21T13:51:34Z

...src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadConstants.java

+
+public class MixedbreadConstants {
+    public static final String VERSION_1 = "v1";
+    public static final String RERANK_PATH = "rerank";


The constant RERANK_PATH is defined as "rerank" here but in MixedbreadRerankModel.java it's defined as "reranking". This inconsistency could lead to incorrect API paths being constructed. These should be unified to use the same value.

Suggested change

public static final String RERANK_PATH = "rerank";

public static final String RERANK_PATH = "reranking";

Mixedbread supports both. I left "rerank". Done

While both may work, the documentation for Mixedbread reranking uses reranking as the endpoint, so that's what we should be using.

jonathan-buttner

Thanks for putting this together, I left some feedback.

jonathan-buttner · 2026-01-21T18:32:27Z

...va/org/elasticsearch/xpack/inference/services/mixedbread/action/MixedbreadActionCreator.java

+    private static final String INVALID_REQUEST_TYPE_MESSAGE = "Invalid request type: expected Mixedbread %s request but got %s";
+
+    private static final ResponseHandler RERANK_HANDLER = new MixedbreadRerankResponseHandler("mixedbread rerank", (request, response) -> {
+        if ((request instanceof MixedbreadRerankRequest) == false) {


I don't believe we typically check the the request type unless we need to use it. If the response format is invalid, the parsing logic will throw an error which should be good enough. Let's remove the if-block.

Thanks. I deleted the if-block

jonathan-buttner · 2026-01-21T18:35:47Z

...va/org/elasticsearch/xpack/inference/services/mixedbread/action/MixedbreadActionCreator.java

+            ),
+            QueryAndDocsInputs.class
+        );
+        var errorMessage = buildErrorMessage(TaskType.RERANK, model.getInferenceEntityId());


Let's use the helper method constructFailedToSendRequestMessage. Take a look at OpenAiActionCreator for example usage.

Cleaned up and used the helper method instead. Thx

jonathan-buttner · 2026-01-21T18:37:05Z

...in/java/org/elasticsearch/xpack/inference/services/mixedbread/request/MixedbreadRequest.java

+    public static void decorateWithAuthHeader(HttpPost request, MixedbreadAccount account) {
+        request.setHeader(HttpHeaders.CONTENT_TYPE, XContentType.JSON.mediaType());
+        request.setHeader(createAuthBearerHeader(account.apiKey()));
+        request.setHeader(new BasicHeader(REQUEST_SOURCE_HEADER, ELASTIC_REQUEST_SOURCE));


Can you point me to documentation as to why we need this header?

I used other implementations as references mainly Cohere, Jina AI and the ones you suggested in the comments.
It happened I ended up with an unnecessary header. Since I didn't find the one to be required.
This class is now deleted due to the changes related to other comments and this header is not used in my implementation.

jonathan-buttner · 2026-01-21T18:43:32Z

...lasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankServiceSettings.java

+     * Write	360	    120	    1-minute
+     * Update	480	    160	    1-minute
+     * Delete	240	    80	    1-minute
+     * <a href="https://www.mixedbread.com/api-reference/rate-limits">Rate Limiting</a>.


These rate limits are for their storage operations. I'm not really sure what that is. If you go to the pricing page we can see that the free tier is limited to 100 requests per minute: https://www.mixedbread.com/pricing

Can you update the value to 100 and add the url I linked?

I updated the url and added the link. Thank you

jonathan-buttner · 2026-01-21T18:58:06Z

...lasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankServiceSettings.java

+    public static MixedbreadRerankServiceSettings fromMap(Map<String, Object> map, ConfigurationParseContext context) {
+        ValidationException validationException = new ValidationException();
+
+        String url = extractOptionalString(map, URL, ModelConfigurations.SERVICE_SETTINGS, validationException);


Does mixedbread allow users to spin up deployments? From poking around it seems like requests are only made to https://api.mixedbread.com. Can we remove this? For testing we'll need a way to pass a local URL. For examples of how to do that take a look at https://github.com/elastic/elasticsearch/blob/main/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/mistral/MistralModel.java

We basically just need a setter on the base model class.

Yep, seems it can be removed. I did it. Also added the the suggested method

jonathan-buttner · 2026-01-21T19:49:50Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+    ) {
+
+    }
+


Also let's override supportsChunkedInfer() and return false.

thrown an exception in doChunkedInfer() and returned false in supportsChunkedInfer()

jonathan-buttner · 2026-01-21T19:50:06Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+    }
+
+    @Override
+    protected void validateInputType(InputType inputType, Model model, ValidationException validationException) {


I think we can remove this.

I removed the implementation from the method

jonathan-buttner · 2026-01-21T19:50:17Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+
+    @Override
+    public TransportVersion getMinimalSupportedVersion() {
+        return TransportVersion.minimumCompatible();


Switch this to use the new style.

jonathan-buttner · 2026-01-21T19:50:49Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+
+    @Override
+    public Set<TaskType> supportedStreamingTasks() {
+        return COMPLETION_ONLY;


We don't support streaming tasks for Mixedbread yet so let's remove this.

jonathan-buttner · 2026-01-21T20:02:47Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+
+    @Override
+    public int rerankerWindowSize(String modelId) {
+        // Cohere rerank model truncates at 4096 tokens https://docs.cohere.com/reference/rerank


This is pointing to cohere, I think we want the numbers posted here: https://www.mixedbread.com/docs/models/reranking/mxbai-rerank-large-v2

Looks like the older models have a window size of 512. We should make this configurable though. Let's add an optional field to the service settings that can control this value and default it to 8k.

jonathan-buttner · 2026-01-21T20:09:45Z

Also please add a change log entry.

…' into Add-Mixedbread-AI-Rerank-support

# Conflicts: # server/src/main/resources/transport/upper_bounds/9.4.csv

Evgenii-Kazannik · 2026-01-29T16:46:08Z

Currently I got one check failed: Elasticsearch Serverless Checks
I'm not sure about this one yet. The link leads to Page not found
Local ./gredlew check s pass successfully

Evgenii-Kazannik · 2026-01-29T16:47:42Z

Regarding the comment. The change log entry is added

# Conflicts: # server/src/main/resources/transport/upper_bounds/9.4.csv

DonalEvans · 2026-01-30T20:18:37Z

docs/changelog/140477.yaml

@@ -0,0 +1,5 @@
+pr: 140477
+summary: "[ML] Add Mixedbread Rerank support to the Inference Plugin"
+area: Machine Learning


The area should be "Inference" rather than "Machine Learning"

Replaced: [ML] -> [Inference API]

That's also a good change, but I was referring to the area field, which should be Inference, not Machine Learning.

DonalEvans · 2026-01-30T20:23:52Z

...a/org/elasticsearch/xpack/inference/services/mixedbread/request/MixedbreadRerankRequest.java

+        this.query = Objects.requireNonNull(query);
+        this.returnDocuments = returnDocuments;
+        this.topN = topN;
+        taskSettings = model.getTaskSettings();


Nitpick, but rather than having a taskSettings field, it would be a little neater and more consistent to just get the task settings from model when we need them, like we do with the model ID and URI.

Good. Thank you. Done

DonalEvans · 2026-01-30T20:41:01Z

...java/org/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankModel.java

+import static org.elasticsearch.xpack.inference.external.request.RequestUtils.buildUri;
+
+public class MixedbreadRerankModel extends MixedbreadModel {
+    private URI uri = buildUri(


Instead of having this be a mutable field, it could be final and set in the constructor using a constant value if the provided argument is null, similar to what's done in the JinaAI*Model classes, for example. This makes the class immutable, but still allows a non-default URL to be passed in for testing purposes.

Did it, also deleted the setter

DonalEvans · 2026-01-30T20:44:25Z

...java/org/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankModel.java

+    }
+
+    // should only be used for testing
+    public MixedbreadRerankModel(


This can be default visibility rather than public, since it's not intended to be used outside this package.

DonalEvans · 2026-01-30T21:00:49Z

...java/org/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankModel.java

+    public MixedbreadRerankModel(MixedbreadRerankModel model, MixedbreadRerankServiceSettings serviceSettings) {
+        super(model, serviceSettings);
+    }


This constructor is unused, do we need it?

No. Deleted. Thx

DonalEvans · 2026-01-30T23:47:32Z

.../test/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadServiceTests.java

+        }
+    }
+
+    public void testInfer_Rerank_Get_Response_NoReturnDocuments_NoTopN() throws IOException {


Do these tests need the "Get_Response" in their names? I'm not sure what it actually means in this context.

The naming came from Jina AI service I used as a reference. We're checking if we get the results from the response we mocked so both having and not having Get_Response in the name is probably okay. I made it simpler

DonalEvans · 2026-01-30T23:50:03Z

.../test/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadServiceTests.java

+                null,
+                null,
+                INPUT,
+                false,


This false should be extracted to a variable along with the corresponding value in the assertion at the end of this test.

DonalEvans · 2026-01-30T23:51:04Z

.../test/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadServiceTests.java

+
+        try (var service = new MixedbreadService(senderFactory, createWithEmptySettings(threadPool), mockClusterServiceEmpty())) {
+            webServer.enqueue(new MockResponse().setResponseCode(200).setBody(responseJson));
+            var model = MixedbreadRerankModelTests.createModel(MODEL_NAME_VALUE, "secret", 3, true);


The 3 and true values should be extracted to variables.

DonalEvans · 2026-01-30T23:55:07Z

.../test/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadServiceTests.java

+    private Map<String, Object> getRequestConfigMap(
+        Map<String, Object> serviceSettings,
+        Map<String, Object> taskSettings,
+        Map<String, Object> secretSettings
+    ) {
+        var builtServiceSettings = new HashMap<>();
+        builtServiceSettings.putAll(serviceSettings);
+        builtServiceSettings.putAll(secretSettings);
+
+        return new HashMap<>(
+            Map.of(ModelConfigurations.SERVICE_SETTINGS, builtServiceSettings, ModelConfigurations.TASK_SETTINGS, taskSettings)
+        );
+    }


An identical method already exists in the Utils test class, so we should use that one instead.

DonalEvans · 2026-01-30T23:57:39Z

.../test/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadServiceTests.java

+    private Map<String, Object> getRequestConfigMap(Map<String, Object> serviceSettings, Map<String, Object> secretSettings) {
+        var builtServiceSettings = new HashMap<>();
+        builtServiceSettings.putAll(serviceSettings);
+        builtServiceSettings.putAll(secretSettings);
+
+        return new HashMap<>(Map.of(ModelConfigurations.SERVICE_SETTINGS, builtServiceSettings));
+    }


Is this method necessary? Would it be possible to just use the three-argument version of Utils.getRequestConfigMap() and pass Map.of() for the second argument (the task settings map)?

Yep, that's better. I re-used the method we got in Utils. Also cleaned up the test class. Thanks

# Conflicts: # server/src/main/resources/transport/upper_bounds/9.4.csv

DonalEvans · 2026-02-02T16:50:21Z

...lasticsearch/xpack/inference/services/mixedbread/request/rerank/MixedbreadRerankRequest.java

        this.returnDocuments = returnDocuments;
        this.topN = topN;
-        taskSettings = model.getTaskSettings();
+        model.getTaskSettings();


This line should be removed.

DonalEvans · 2026-02-02T17:11:32Z

...e/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadService.java

+
+    @Override
+    public int rerankerWindowSize(String modelId) {
+        // Cohere rerank model truncates at 4096 tokens https://docs.cohere.com/reference/rerank


The window size is not something a user can configure, it's an unchanging property of the model, so we don't need to store it in service settings. The current approach of having a map with the model IDs that don't use the 8k default is fine, but the rerankerWindowSize() method returns the size in words, not in tokens, so we'll need to translate from the 512/8000 values in tokens to smaller values in words by multiplying by 0.75 and rounding down a bit, which is the approach we use for other providers.

For consistency, the PR that originally introduced this feature can be used as a guide, with 512 tokens translating to a window size of 300 words, and 8000 tokens translating to 5500 words.

DonalEvans · 2026-02-02T17:13:48Z

...g/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankTaskSettings.java

-            ModelConfigurations.TASK_SETTINGS,
-            validationException
-        );
+        Integer topNDocumentsOnly = extractOptionalPositiveInteger(map, TOP_N, ModelConfigurations.TASK_SETTINGS, validationException);


This variable name can also be updated to remove the "DocumentsOnly" part, since it doesn't seem to have any relevance here.

DonalEvans · 2026-02-02T17:19:56Z

...asticsearch/xpack/inference/services/mixedbread/response/MixedbreadRerankResponseEntity.java

     * }
     * <p>
     *  The response will look like (without whitespace):
+     *  <pre>


The request JSON example should also have <pre> tags to make it readable when rendered by the IDE or as HTML.

DonalEvans · 2026-02-02T17:20:40Z

...nce/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadModel.java

        @Nullable ApiKeySecrets apiKeySecrets,
-        RateLimitSettings rateLimitServiceSettings
+        RateLimitSettings rateLimitServiceSettings,
+        URI uri


The uri field should be made private final instead of protected.

DonalEvans · 2026-02-02T17:22:39Z

...nce/src/main/java/org/elasticsearch/xpack/inference/services/mixedbread/MixedbreadUtils.java

     * TransportVersion indicating when Mixedbread features were added.
     */
-    public static final TransportVersion ML_INFERENCE_MIXEDBREAD_ADDED = TransportVersion.fromName("ml_inference_mixedbread_added");
+    public static final TransportVersion INFERENCE_MIXEDBREAD_ADDED = TransportVersion.fromName("ml_inference_mixedbread_added");


Nitpick, but can we remove the "ml" from the start of the TransportVersion name?

DonalEvans · 2026-02-02T17:42:31Z

...org/elasticsearch/xpack/inference/services/mixedbread/rerank/MixedbreadRerankModelTests.java

        @Nullable Integer topN,
-        @Nullable Boolean returnDocuments
+        @Nullable Boolean returnDocuments,
+        String uri


This should be annotated with @Nullable

DonalEvans · 2026-02-02T17:59:03Z

Sorry, I forgot to include this in the previous review, but one other thing that should be added to this PR are some tests of the behaviour when there are unexpected fields in the maps passed to parseRequestConfig(), parsePersistedConfig() and parsePersistedConfigWithSecrets() in MixedbreadServiceTests. Unexpected fields in requests should result in exceptions, and unexpected fields in persisted configs should be ignored.

elasticsearchmachine added v9.4.0 external-contributor Pull request authored by a developer outside the Elasticsearch team labels Jan 9, 2026

Evgenii-Kazannik added 4 commits January 13, 2026 15:40

Add Mixedbread AI Rerank support

f52f5e2

Add Mixedbread AI Rerank support tests

0a184ca

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

dc1f701

Apply spotless

6133d64

Evgenii-Kazannik force-pushed the Add-Mixedbread-AI-Rerank-support branch from 2683926 to 6133d64 Compare January 13, 2026 14:56

Evgenii-Kazannik added external-contributor Pull request authored by a developer outside the Elasticsearch team and removed external-contributor Pull request authored by a developer outside the Elasticsearch team labels Jan 14, 2026

Evgenii-Kazannik added 2 commits January 14, 2026 18:17

Add Mixedbread AI Rerank support

6b63ffb

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

2cd922c

Evgenii-Kazannik marked this pull request as ready for review January 14, 2026 17:24

elasticsearchmachine added the needs:triage Requires assignment of a team area label label Jan 14, 2026

Evgenii-Kazannik added 2 commits January 20, 2026 15:39

Add action creator tests

9ec0f7d

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

3fd0d23

jonathan-buttner self-assigned this Jan 20, 2026

jonathan-buttner added :SearchOrg/Inference Label for the Search Inference team Team:Search - Inference >enhancement and removed needs:triage Requires assignment of a team area label labels Jan 20, 2026

jonathan-buttner requested a review from Copilot January 21, 2026 13:40

Copilot started reviewing on behalf of jonathan-buttner January 21, 2026 13:41 View session

Copilot AI reviewed Jan 21, 2026

View reviewed changes

jonathan-buttner requested changes Jan 21, 2026

View reviewed changes

Evgenii-Kazannik and others added 4 commits January 22, 2026 21:06

Make windows size configurable

4cb12db

Address comments and add service tests

18d5ce4

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

fd98ef0

[CI] Update transport version definitions

5bbffe1

Evgenii-Kazannik and others added 12 commits January 28, 2026 05:26

Switch to new approach for transport version

83a6497

Use ConstructingObjectParser

fbd6d5d

Address comments

f4c5c0d

Merge remote-tracking branch 'origin/Add-Mixedbread-AI-Rerank-support…

c72e6a2

…' into Add-Mixedbread-AI-Rerank-support

[CI] Auto commit changes from spotless

faa5ce8

Fix the test

36f7994

Checkstyle fix

87f09e5

Fix the test

fa208ee

Clean up

5ed8e6a

Clean up

6c3bc8f

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

cd94963

# Conflicts: # server/src/main/resources/transport/upper_bounds/9.4.csv

[CI] Update transport version definitions

81f9aca

Evgenii-Kazannik and others added 3 commits January 29, 2026 17:59

ci: retrigger

e447752

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

59d97d8

# Conflicts: # server/src/main/resources/transport/upper_bounds/9.4.csv

[CI] Update transport version definitions

68108cf

DonalEvans requested changes Jan 30, 2026

View reviewed changes

Evgenii-Kazannik and others added 3 commits February 2, 2026 14:30

Address comments and refactor

2ab34e0

Merge branch 'main' into Add-Mixedbread-AI-Rerank-support

65f2b64

# Conflicts: # server/src/main/resources/transport/upper_bounds/9.4.csv

[CI] Update transport version definitions

a20144f

DonalEvans reviewed Feb 2, 2026

View reviewed changes

Evgenii-Kazannik and others added 2 commits February 3, 2026 01:24

Address comments

9b1a82c

[CI] Update transport version definitions

84aab0e

	public void UpdatedTaskSettings_WithEmptyMap_ReturnsSameSettings() {
	public void testUpdatedTaskSettings_WithEmptyMap_ReturnsSameSettings() {


		import java.util.Map;

		import static org.elasticsearch.xpack.inference.services.jinaai.rerank.JinaAIRerankTaskSettingsTests.getTaskSettingsMap;

	positionParserAtTokenAfterField(parser, "data", "FAILED_TO_FIND_FIELD_TEMPLATE"); // TODO error message
	positionParserAtTokenAfterField(parser, "data", "Failed to find [data] field in Mixedbread rerank response");

	// Cohere rerank model truncates at 4096 tokens https://docs.cohere.com/reference/rerank
	// Mixedbread rerank models currently support context windows of up to 4096 tokens (see Mixedbread documentation)


	throw new UnsupportedOperationException("Chunked inference is not supported for service [" + NAME + "]");

	public static final String RERANK_PATH = "rerank";
	public static final String RERANK_PATH = "reranking";

Add Mixedbread AI Rerank support #140477

Are you sure you want to change the base?

Add Mixedbread AI Rerank support #140477

Conversation

Evgenii-Kazannik commented Jan 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

elasticsearchmachine commented Jan 20, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Evgenii-Kazannik Jan 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Copilot AI Jan 21, 2026

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jonathan-buttner left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Evgenii-Kazannik commented Jan 9, 2026 •

edited

Loading

Evgenii-Kazannik Jan 27, 2026 •

edited

Loading