diff --git a/.github/workflows/sync-models.yml b/.github/workflows/sync-models.yml
index c46c20519a..8f5434f7c0 100644
--- a/.github/workflows/sync-models.yml
+++ b/.github/workflows/sync-models.yml
@@ -65,6 +65,7 @@ jobs:
         env:
           BASETEN_API_KEY: ${{ secrets.BASETEN_API_KEY }}
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
+          NEBIUS_API_KEY: ${{ secrets.NEBIUS_API_KEY }}
           OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
           VENICE_API_KEY: ${{ secrets.VENICE_API_KEY }}
           LLMGATEWAY_API_KEY: ${{ secrets.LLMGATEWAY_API_KEY }}
diff --git a/packages/core/src/sync/index.ts b/packages/core/src/sync/index.ts
index 3229b15679..8343af3d51 100644
--- a/packages/core/src/sync/index.ts
+++ b/packages/core/src/sync/index.ts
@@ -10,6 +10,7 @@ import { cloudflareWorkersAi } from "./providers/cloudflare-workers-ai.js";
 import { google } from "./providers/google.js";
 import { huggingface } from "./providers/huggingface.js";
 import { llmgateway } from "./providers/llmgateway.js";
+import { nebius } from "./providers/nebius.js";
 import { openrouter } from "./providers/openrouter.js";
 import { ovhcloud } from "./providers/ovhcloud.js";
 import { vercel } from "./providers/vercel.js";
@@ -86,6 +87,7 @@ export const providers: {
   google: SyncProvider<any>;
   huggingface: SyncProvider<any>;
   llmgateway: SyncProvider<any>;
+  nebius: SyncProvider<any>;
   openrouter: SyncProvider<any>;
   ovhcloud: SyncProvider<any>;
   vercel: SyncProvider<any>;
@@ -98,6 +100,7 @@ export const providers: {
   google,
   huggingface,
   llmgateway,
+  nebius,
   openrouter,
   ovhcloud,
   vercel,
@@ -108,7 +111,7 @@ export const providers: {
 export const groups = {
   aggregators: ["huggingface", "llmgateway", "openrouter", "vercel"],
   cloudflare: ["cloudflare-workers-ai"],
-  direct: ["baseten", "chutes", "google", "ovhcloud", "venice", "xai"],
+  direct: ["baseten", "chutes", "google", "nebius", "ovhcloud", "venice", "xai"],
 } as const;
 
 type ProviderID = keyof typeof providers;
diff --git a/packages/core/src/sync/providers/nebius.ts b/packages/core/src/sync/providers/nebius.ts
new file mode 100644
index 0000000000..c10a69f19e
--- /dev/null
+++ b/packages/core/src/sync/providers/nebius.ts
@@ -0,0 +1,262 @@
+import { readdirSync } from "node:fs";
+import path from "node:path";
+import { z } from "zod";
+
+import { inferKimiFamily, ModelFamilyValues } from "../../family.js";
+import type { ExistingModel, SyncProvider, SyncedFullModel, SyncedModel } from "../index.js";
+import { factorBaseModel } from "./openrouter.js";
+
+const API_ENDPOINT = "https://api.tokenfactory.nebius.com/v1/models";
+const MODELS_DIR = path.join(import.meta.dirname, "..", "..", "..", "..", "..", "models");
+
+// Scoped to the orgs Nebius Token Factory currently hosts; extend as new orgs show up.
+// Maps a Nebius org prefix to the provider-agnostic namespace under `models/`.
+const NEBIUS_ORG_TO_MODEL_PROVIDER: Record<string, string | undefined> = {
+  "meta-llama": "meta",
+  Qwen: "alibaba",
+  google: "google",
+  openai: "openai",
+  nvidia: "nvidia",
+  "zai-org": "zhipuai",
+  moonshotai: "moonshotai",
+  MiniMaxAI: "minimax",
+  "deepseek-ai": "deepseek",
+};
+
+const modelMetadataFilesByProvider = new Map<string, Set<string>>();
+
+const NebiusPricing = z.object({
+  prompt: z.string().optional(),
+  completion: z.string().optional(),
+}).passthrough();
+
+const NebiusArchitecture = z.object({
+  modality: z.string(),
+  tokenizer: z.string().optional(),
+}).passthrough();
+
+export const NebiusModel = z.object({
+  id: z.string(),
+  name: z.string().optional(),
+  created: z.number().optional(),
+  context_length: z.number().optional(),
+  architecture: NebiusArchitecture.optional(),
+  pricing: NebiusPricing.optional(),
+  supported_features: z.array(z.string()).optional(),
+  supported_sampling_parameters: z.array(z.string()).optional(),
+}).passthrough();
+
+export const NebiusResponse = z.object({
+  data: z.array(NebiusModel),
+}).passthrough();
+
+export type NebiusModel = z.infer<typeof NebiusModel>;
+
+type Modality = "text" | "audio" | "image" | "video" | "pdf";
+
+const KNOWN_MODALITIES = new Set<Modality>(["text", "audio", "image", "video", "pdf"]);
+
+export const nebius = {
+  id: "nebius",
+  name: "Nebius Token Factory",
+  modelsDir: "providers/nebius/models",
+  async fetchModels() {
+    const key = process.env.NEBIUS_API_KEY;
+    if (key === undefined) throw new Error("Nebius sync requires NEBIUS_API_KEY");
+
+    const url = new URL(API_ENDPOINT);
+    // `verbose=true` is required for pricing, context length, modality, and feature
+    // metadata; the default response only includes bare model IDs.
+    url.searchParams.set("verbose", "true");
+
+    const response = await fetch(url, {
+      headers: { Authorization: `Bearer ${key}` },
+    });
+    if (!response.ok) {
+      throw new Error(`Nebius models request failed: ${response.status} ${response.statusText}`);
+    }
+    return response.json();
+  },
+  parseModels(raw) {
+    return NebiusResponse.parse(raw).data;
+  },
+  translateModel(model, context) {
+    return {
+      id: model.id,
+      model: buildNebiusModel(model, context.existing(model.id)),
+    };
+  },
+} satisfies SyncProvider<NebiusModel>;
+
+export function buildNebiusModel(
+  model: NebiusModel,
+  existing: ExistingModel | undefined,
+  today = new Date().toISOString().slice(0, 10),
+): SyncedModel {
+  const features = new Set(model.supported_features ?? []);
+  const samplingParameters = new Set(model.supported_sampling_parameters ?? []);
+  const { input, output } = parseModality(model.architecture?.modality, existing?.modalities);
+  const name = modelName(model, existing);
+  const attachment = input.some((value) => value !== "text");
+  const reasoning = features.has("reasoning");
+  const toolCall = features.has("tools");
+  const structuredOutput = features.has("structured_outputs");
+  // Nebius omits `supported_sampling_parameters` entirely for some model families;
+  // an empty list is ambiguous, so fall back to the existing flag in that case.
+  const temperature = samplingParameters.size > 0
+    ? samplingParameters.has("temperature")
+    : existing?.temperature ?? true;
+  const context = model.context_length ?? existing?.limit?.context ?? 0;
+  const limit = {
+    context,
+    input: existing?.limit?.input,
+    output: existing?.limit?.output,
+  };
+  const releaseDate = existing?.release_date ?? today;
+  const canonical = existing?.base_model ?? resolveCanonicalBaseModel(model.id);
+
+  if (canonical !== undefined) {
+    return factorBaseModel(
+      canonical,
+      {
+        name: existing === undefined ? name : undefined,
+        attachment,
+        reasoning,
+        // The API only exposes a boolean `reasoning` capability, not the toggle/effort
+        // mechanism, so any reasoning_options are curated by hand and preserved as-is.
+        reasoning_options: existing?.reasoning_options,
+        temperature,
+        tool_call: toolCall,
+        structured_output: structuredOutput,
+        status: existing?.status,
+        interleaved: existing?.interleaved,
+        limit,
+        modalities: { input, output },
+        cost: cost(model, existing),
+      },
+      limit,
+      existing?.base_model === canonical ? existing.base_model_omit : undefined,
+    );
+  }
+
+  return {
+    name,
+    family: existing?.family ?? inferFamily(model.id, name),
+    release_date: releaseDate,
+    last_updated: existing?.last_updated ?? today,
+    attachment,
+    reasoning,
+    reasoning_options: existing?.reasoning_options,
+    temperature,
+    tool_call: toolCall,
+    structured_output: structuredOutput,
+    knowledge: existing?.knowledge,
+    open_weights: existing?.open_weights ?? true,
+    status: existing?.status,
+    interleaved: existing?.interleaved,
+    cost: cost(model, existing),
+    limit: {
+      input: limit.input,
+      context: limit.context,
+      // No authoritative output limit is available without canonical metadata;
+      // fall back to the served context length, same as OpenRouter's convention.
+      output: limit.output ?? context,
+    },
+    modalities: { input, output },
+  } satisfies SyncedFullModel;
+}
+
+function modelName(model: NebiusModel, existing: ExistingModel | undefined) {
+  // Nebius occasionally bakes the org prefix into `name` (e.g. "openbmb/MiniCPM-V-4_5");
+  // strip any such prefix rather than let it leak into the catalog.
+  const raw = model.name ?? existing?.name ?? model.id;
+  return raw.split("/").at(-1) ?? raw;
+}
+
+function price(value: string | undefined) {
+  if (value === undefined) return undefined;
+  const number = Number(value);
+  return Number.isFinite(number) && number >= 0
+    ? Math.round(number * 1_000_000_000_000) / 1_000_000
+    : undefined;
+}
+
+function cost(model: NebiusModel, existing: ExistingModel | undefined) {
+  const input = price(model.pricing?.prompt);
+  const output = price(model.pricing?.completion);
+  if (input === undefined || output === undefined) return existing?.cost;
+
+  return {
+    input,
+    output,
+    // Cache pricing is not exposed by the Models API; existing ratios are curated by hand.
+    reasoning: existing?.cost?.reasoning,
+    cache_read: existing?.cost?.cache_read,
+    cache_write: existing?.cost?.cache_write,
+    tiers: existing?.cost?.tiers,
+  };
+}
+
+function parseModality(
+  modality: string | undefined,
+  fallback: ExistingModel["modalities"],
+) {
+  const [rawInput, rawOutput] = (modality ?? "text->text").split("->");
+  return {
+    input: parseModalityPart(rawInput, fallback?.input ?? ["text"]),
+    output: parseModalityPart(rawOutput, fallback?.output ?? ["text"]),
+  };
+}
+
+function parseModalityPart(part: string | undefined, fallback: Modality[]): Modality[] {
+  const values = (part ?? "")
+    .split("+")
+    .map((value) => value.trim().toLowerCase())
+    // Nebius reports embedding models as `text->embedding`; the catalog schema has no
+    // embedding modality, and embedding models are catalogued as text-in/text-out.
+    .map((value) => (value === "embedding" ? "text" : value))
+    .filter((value): value is Modality => KNOWN_MODALITIES.has(value as Modality));
+
+  return [...new Set(values.length > 0 ? values : fallback)];
+}
+
+function inferFamily(id: string, name: string) {
+  const kimiFamily = inferKimiFamily(id, name);
+  if (kimiFamily !== undefined) return kimiFamily;
+
+  const target = `${id} ${name}`.toLowerCase();
+  return [...ModelFamilyValues]
+    .sort((a, b) => b.length - a.length)
+    .find((family) => {
+      const value = family.toLowerCase().replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+      if (family === "o") return new RegExp(`(^|[^a-z0-9])${value}(?=\\d|$|[^a-z0-9])`).test(target);
+      return new RegExp(`(^|[^a-z0-9])${value}(?=$|[^a-z0-9])`).test(target);
+    });
+}
+
+export function resolveCanonicalBaseModel(nebiusID: string): string | undefined {
+  const [org, ...modelParts] = nebiusID.split("/");
+  if (org === undefined || modelParts.length === 0) return undefined;
+
+  const provider = NEBIUS_ORG_TO_MODEL_PROVIDER[org];
+  if (provider === undefined) return undefined;
+
+  const modelID = modelParts.join("/");
+  const candidates = [...new Set([modelID, modelID.toLowerCase()])];
+  const match = candidates.find((candidate) => modelMetadataExists(provider, candidate));
+
+  return match === undefined ? undefined : `${provider}/${match}`;
+}
+
+function modelMetadataExists(provider: string, modelID: string) {
+  let files = modelMetadataFilesByProvider.get(provider);
+  if (files === undefined) {
+    try {
+      files = new Set(readdirSync(path.join(MODELS_DIR, provider)));
+    } catch {
+      files = new Set();
+    }
+    modelMetadataFilesByProvider.set(provider, files);
+  }
+  return files.has(`${modelID}.toml`);
+}
diff --git a/providers/nebius/models/MiniMaxAI/MiniMax-M2.5-fast.toml b/providers/nebius/models/MiniMaxAI/MiniMax-M2.5-fast.toml
deleted file mode 100644
index e4d44d35b6..0000000000
--- a/providers/nebius/models/MiniMaxAI/MiniMax-M2.5-fast.toml
+++ /dev/null
@@ -1,27 +0,0 @@
-name = "MiniMax-M2.5-fast"
-attachment = false
-reasoning = true
-reasoning_options = []
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-01"
-release_date = "2025-01-20"
-last_updated = "2026-05-07"
-open_weights = true
-
-[cost]
-input = 0.30
-output = 1.20
-cache_read = 0.03
-cache_write = 0.375
-
-[limit]
-context = 8_000
-input = 7_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/MiniMaxAI/MiniMax-M2.5.toml b/providers/nebius/models/MiniMaxAI/MiniMax-M2.5.toml
index 448e398d94..59d1616537 100644
--- a/providers/nebius/models/MiniMaxAI/MiniMax-M2.5.toml
+++ b/providers/nebius/models/MiniMaxAI/MiniMax-M2.5.toml
@@ -1,26 +1,14 @@
-name = "MiniMax-M2.5"
-attachment = false
-reasoning = true
+base_model = "minimax/MiniMax-M2.5"
+structured_output = false
 reasoning_options = []
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-01"
-release_date = "2025-01-20"
-last_updated = "2026-05-07"
-open_weights = true
 
 [cost]
-input = 0.30
-output = 1.20
+input = 0.3
+output = 1.2
 cache_read = 0.03
 cache_write = 0.375
 
 [limit]
-context = 196_608
+context = 8_000
 input = 190_000
 output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/NousResearch/Hermes-4-405B.toml b/providers/nebius/models/NousResearch/Hermes-4-405B.toml
index 61d1275cd9..6fe70b3c4b 100644
--- a/providers/nebius/models/NousResearch/Hermes-4-405B.toml
+++ b/providers/nebius/models/NousResearch/Hermes-4-405B.toml
@@ -1,30 +1,31 @@
 name = "Hermes-4-405B"
+family = "nousresearch"
+release_date = "2026-01-30"
+last_updated = "2026-02-04"
 attachment = false
 reasoning = true
-reasoning_options = []
+temperature = true
 tool_call = true
 structured_output = true
-temperature = true
 knowledge = "2025-11"
-release_date = "2026-01-30"
-last_updated = "2026-02-04"
 open_weights = true
+reasoning_options = []
+
+[interleaved]
+field = "reasoning_content"
 
 [cost]
-input = 1.00
-output = 3.00
-reasoning = 3.00
-cache_read = 0.10
+input = 1
+output = 3
+reasoning = 3
+cache_read = 0.1
 cache_write = 1.25
 
 [limit]
-context = 128_000
+context = 131_072
 input = 120_000
 output = 8_192
 
 [modalities]
 input = ["text"]
 output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/NousResearch/Hermes-4-70B.toml b/providers/nebius/models/NousResearch/Hermes-4-70B.toml
index 673a4fdce7..849c6aab06 100644
--- a/providers/nebius/models/NousResearch/Hermes-4-70B.toml
+++ b/providers/nebius/models/NousResearch/Hermes-4-70B.toml
@@ -1,30 +1,31 @@
 name = "Hermes-4-70B"
+family = "nousresearch"
+release_date = "2026-01-30"
+last_updated = "2026-02-04"
 attachment = false
 reasoning = true
-reasoning_options = []
+temperature = true
 tool_call = true
 structured_output = true
-temperature = true
 knowledge = "2025-11"
-release_date = "2026-01-30"
-last_updated = "2026-02-04"
 open_weights = true
+reasoning_options = []
+
+[interleaved]
+field = "reasoning_content"
 
 [cost]
 input = 0.13
-output = 0.40
-reasoning = 0.40
+output = 0.4
+reasoning = 0.4
 cache_read = 0.013
 cache_write = 0.16
 
 [limit]
-context = 128_000
+context = 131_072
 input = 120_000
 output = 8_192
 
 [modalities]
 input = ["text"]
 output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/PrimeIntellect/INTELLECT-3.toml b/providers/nebius/models/PrimeIntellect/INTELLECT-3.toml
deleted file mode 100644
index d2de6f510b..0000000000
--- a/providers/nebius/models/PrimeIntellect/INTELLECT-3.toml
+++ /dev/null
@@ -1,26 +0,0 @@
-name = "INTELLECT-3"
-attachment = false
-reasoning = false
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-10"
-release_date = "2026-01-25"
-last_updated = "2026-02-04"
-open_weights = true
-
-[cost]
-input = 0.20
-output = 1.10
-cache_read = 0.02
-cache_write = 0.25
-
-[limit]
-context = 128_000
-input = 120_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen2.5-VL-72B-Instruct.toml b/providers/nebius/models/Qwen/Qwen2.5-VL-72B-Instruct.toml
index b21f2c926d..cbe13a9bb1 100644
--- a/providers/nebius/models/Qwen/Qwen2.5-VL-72B-Instruct.toml
+++ b/providers/nebius/models/Qwen/Qwen2.5-VL-72B-Instruct.toml
@@ -1,12 +1,13 @@
 name = "Qwen2.5-VL-72B-Instruct"
+family = "qwen"
+release_date = "2025-01-20"
+last_updated = "2026-02-04"
 attachment = true
 reasoning = false
-tool_call = true
-structured_output = true
 temperature = true
+tool_call = false
+structured_output = false
 knowledge = "2024-12"
-release_date = "2025-01-20"
-last_updated = "2026-02-04"
 open_weights = true
 
 [cost]
@@ -16,10 +17,10 @@ cache_read = 0.025
 cache_write = 0.31
 
 [limit]
-context = 128_000
+context = 32_000
 input = 120_000
 output = 8_192
 
 [modalities]
 input = ["text", "image"]
-output = ["text"]
\ No newline at end of file
+output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml b/providers/nebius/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml
index 7ced37cf17..7b5066c07c 100644
--- a/providers/nebius/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml
+++ b/providers/nebius/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml
@@ -1,21 +1,22 @@
-name = "Qwen3 235B A22B Instruct 2507"
+name = "Qwen3-235B-A22B-Instruct-2507"
 family = "qwen"
 release_date = "2025-07-25"
 last_updated = "2025-10-04"
 attachment = false
 reasoning = false
 temperature = true
-knowledge = "2025-07"
 tool_call = true
+structured_output = true
+knowledge = "2025-07"
 open_weights = false
 
 [cost]
-input = 0.20
-output = 0.60
+input = 0.2
+output = 0.6
 
 [limit]
-context = 262144
-output = 8192
+context = 262_144
+output = 8_192
 
 [modalities]
 input = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3-235B-A22B-Thinking-2507-fast.toml b/providers/nebius/models/Qwen/Qwen3-235B-A22B-Thinking-2507-fast.toml
deleted file mode 100644
index 68fb846cd3..0000000000
--- a/providers/nebius/models/Qwen/Qwen3-235B-A22B-Thinking-2507-fast.toml
+++ /dev/null
@@ -1,27 +0,0 @@
-name = "Qwen3-235B-A22B-Thinking-2507-fast"
-attachment = false
-reasoning = true
-reasoning_options = []
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-07"
-release_date = "2025-07-25"
-last_updated = "2026-05-07"
-open_weights = true
-
-[cost]
-input = 0.50
-output = 2.00
-cache_read = 0.05
-cache_write = 0.625
-
-[limit]
-context = 8_000
-input = 7_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3-30B-A3B-Instruct-2507.toml b/providers/nebius/models/Qwen/Qwen3-30B-A3B-Instruct-2507.toml
index 96c2d3f369..94d05d85dd 100644
--- a/providers/nebius/models/Qwen/Qwen3-30B-A3B-Instruct-2507.toml
+++ b/providers/nebius/models/Qwen/Qwen3-30B-A3B-Instruct-2507.toml
@@ -1,25 +1,26 @@
 name = "Qwen3-30B-A3B-Instruct-2507"
+family = "qwen"
+release_date = "2026-01-28"
+last_updated = "2026-02-04"
 attachment = false
 reasoning = false
+temperature = true
 tool_call = true
 structured_output = true
-temperature = true
 knowledge = "2025-12"
-release_date = "2026-01-28"
-last_updated = "2026-02-04"
 open_weights = true
 
 [cost]
-input = 0.10
-output = 0.30
+input = 0.1
+output = 0.3
 cache_read = 0.01
 cache_write = 0.125
 
 [limit]
-context = 128_000
+context = 262_144
 input = 120_000
 output = 8_192
 
 [modalities]
 input = ["text"]
-output = ["text"]
\ No newline at end of file
+output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3-32B.toml b/providers/nebius/models/Qwen/Qwen3-32B.toml
index 6c619f6e5b..e1f40bd311 100644
--- a/providers/nebius/models/Qwen/Qwen3-32B.toml
+++ b/providers/nebius/models/Qwen/Qwen3-32B.toml
@@ -1,25 +1,14 @@
-name = "Qwen3-32B"
-attachment = false
-reasoning = false
-tool_call = true
+base_model = "alibaba/qwen3-32b"
 structured_output = true
-temperature = true
-knowledge = "2025-12"
-release_date = "2026-01-28"
-last_updated = "2026-02-04"
-open_weights = true
+reasoning_options = []
 
 [cost]
-input = 0.10
-output = 0.30
+input = 0.1
+output = 0.3
 cache_read = 0.01
 cache_write = 0.125
 
 [limit]
-context = 128_000
+context = 40_960
 input = 120_000
 output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
\ No newline at end of file
diff --git a/providers/nebius/models/Qwen/Qwen3-Embedding-8B.toml b/providers/nebius/models/Qwen/Qwen3-Embedding-8B.toml
index 205cee1bc9..234ac30a71 100644
--- a/providers/nebius/models/Qwen/Qwen3-Embedding-8B.toml
+++ b/providers/nebius/models/Qwen/Qwen3-Embedding-8B.toml
@@ -1,24 +1,24 @@
 name = "Qwen3-Embedding-8B"
 family = "text-embedding"
+release_date = "2026-01-10"
+last_updated = "2026-02-04"
 attachment = false
 reasoning = false
+temperature = false
 tool_call = false
 structured_output = false
-temperature = false
 knowledge = "2025-10"
-release_date = "2026-01-10"
-last_updated = "2026-02-04"
 open_weights = true
 
 [cost]
 input = 0.01
-output = 0.00
+output = 0
 
 [limit]
-context = 32_768
+context = 40_960
 input = 32_768
 output = 0
 
 [modalities]
 input = ["text"]
-output = ["text"]
\ No newline at end of file
+output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking-fast.toml b/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking-fast.toml
deleted file mode 100644
index 20b8b16cf6..0000000000
--- a/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking-fast.toml
+++ /dev/null
@@ -1,27 +0,0 @@
-name = "Qwen3-Next-80B-A3B-Thinking-fast"
-attachment = false
-reasoning = true
-reasoning_options = []
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-07"
-release_date = "2025-07-25"
-last_updated = "2026-05-07"
-open_weights = true
-
-[cost]
-input = 0.15
-output = 1.20
-cache_read = 0.015
-cache_write = 0.1875
-
-[limit]
-context = 8_000
-input = 7_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking.toml b/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking.toml
index 926f404dfb..3d0be3a65b 100644
--- a/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking.toml
+++ b/providers/nebius/models/Qwen/Qwen3-Next-80B-A3B-Thinking.toml
@@ -1,30 +1,18 @@
-name = "Qwen3-Next-80B-A3B-Thinking"
-attachment = false
-reasoning = true
-reasoning_options = []
-tool_call = true
+base_model = "alibaba/qwen3-next-80b-a3b-thinking"
 structured_output = true
-temperature = true
-knowledge = "2025-12"
-release_date = "2026-01-28"
-last_updated = "2026-02-04"
-open_weights = true
+reasoning_options = []
+
+[interleaved]
+field = "reasoning_content"
 
 [cost]
 input = 0.15
-output = 1.20
-reasoning = 1.20
+output = 1.2
+reasoning = 1.2
 cache_read = 0.015
 cache_write = 0.18
 
 [limit]
-context = 128_000
+context = 8_000
 input = 120_000
 output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/Qwen/Qwen3.5-397B-A17B-fast.toml b/providers/nebius/models/Qwen/Qwen3.5-397B-A17B-fast.toml
deleted file mode 100644
index 0bb3885a9f..0000000000
--- a/providers/nebius/models/Qwen/Qwen3.5-397B-A17B-fast.toml
+++ /dev/null
@@ -1,27 +0,0 @@
-name = "Qwen3.5-397B-A17B-fast"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-07"
-release_date = "2025-07-15"
-last_updated = "2026-05-07"
-open_weights = true
-
-[cost]
-input = 0.60
-output = 3.60
-cache_read = 0.06
-cache_write = 0.75
-
-[limit]
-context = 8_000
-input = 7_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/Qwen/Qwen3.5-397B-A17B.toml b/providers/nebius/models/Qwen/Qwen3.5-397B-A17B.toml
index d007e174e8..2cb9100fc3 100644
--- a/providers/nebius/models/Qwen/Qwen3.5-397B-A17B.toml
+++ b/providers/nebius/models/Qwen/Qwen3.5-397B-A17B.toml
@@ -1,26 +1,19 @@
-name = "Qwen3.5-397B-A17B"
+base_model = "alibaba/qwen3.5-397b-a17b"
 attachment = false
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-07"
-release_date = "2025-07-15"
-last_updated = "2026-05-07"
-open_weights = true
+
+[[reasoning_options]]
+type = "toggle"
 
 [cost]
-input = 0.60
-output = 3.60
+input = 0.6
+output = 3.6
 cache_read = 0.06
 cache_write = 0.75
 
 [limit]
-context = 262_144
+context = 8_000
 input = 250_000
 output = 8_192
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/deepseek-ai/DeepSeek-V3.2-fast.toml b/providers/nebius/models/deepseek-ai/DeepSeek-V3.2-fast.toml
deleted file mode 100644
index a5679059e0..0000000000
--- a/providers/nebius/models/deepseek-ai/DeepSeek-V3.2-fast.toml
+++ /dev/null
@@ -1,27 +0,0 @@
-name = "DeepSeek-V3.2-fast"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-01"
-release_date = "2025-01-27"
-last_updated = "2026-05-07"
-open_weights = true
-
-[cost]
-input = 0.40
-output = 2.00
-cache_read = 0.04
-cache_write = 0.5
-
-[limit]
-context = 8_000
-input = 7_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/deepseek-ai/DeepSeek-V3.2.toml b/providers/nebius/models/deepseek-ai/DeepSeek-V3.2.toml
deleted file mode 100644
index e0dd51c804..0000000000
--- a/providers/nebius/models/deepseek-ai/DeepSeek-V3.2.toml
+++ /dev/null
@@ -1,31 +0,0 @@
-name = "DeepSeek-V3.2"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-11"
-release_date = "2026-01-20"
-last_updated = "2026-02-04"
-open_weights = true
-
-[cost]
-input = 0.30
-output = 0.45
-reasoning = 0.45
-cache_read = 0.03
-cache_write = 0.375
-
-[limit]
-context = 163_000
-input = 160_000
-output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/deepseek-ai/DeepSeek-V4-Pro.toml b/providers/nebius/models/deepseek-ai/DeepSeek-V4-Pro.toml
index 948f4d822d..bd09a38c39 100644
--- a/providers/nebius/models/deepseek-ai/DeepSeek-V4-Pro.toml
+++ b/providers/nebius/models/deepseek-ai/DeepSeek-V4-Pro.toml
@@ -1,13 +1,19 @@
 base_model = "deepseek/deepseek-v4-pro"
-reasoning_options = [
-  { type = "toggle" },
-  { type = "effort", values = ["low", "medium", "high"] },
-]
 
 [interleaved]
 field = "reasoning_content"
 
+[[reasoning_options]]
+type = "toggle"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
+
 [cost]
 input = 1.75
 output = 3.5
 cache_read = 0.15
+
+[limit]
+context = 1_048_576
diff --git a/providers/nebius/models/google/gemma-3-27b-it.toml b/providers/nebius/models/google/gemma-3-27b-it.toml
index f0fc3dbdad..57ea6f279b 100644
--- a/providers/nebius/models/google/gemma-3-27b-it.toml
+++ b/providers/nebius/models/google/gemma-3-27b-it.toml
@@ -1,17 +1,18 @@
 name = "Gemma-3-27b-it"
-attachment = true
+family = "gemma"
+release_date = "2026-01-20"
+last_updated = "2026-02-04"
+attachment = false
 reasoning = false
-tool_call = true
-structured_output = true
 temperature = true
+tool_call = true
+structured_output = false
 knowledge = "2025-10"
-release_date = "2026-01-20"
-last_updated = "2026-02-04"
 open_weights = true
 
 [cost]
-input = 0.10
-output = 0.30
+input = 0.1
+output = 0.3
 cache_read = 0.01
 cache_write = 0.125
 
@@ -21,5 +22,5 @@ input = 100_000
 output = 8_192
 
 [modalities]
-input = ["text", "image"]
-output = ["text"]
\ No newline at end of file
+input = ["text"]
+output = ["text"]
diff --git a/providers/nebius/models/meta-llama/Llama-3.3-70B-Instruct.toml b/providers/nebius/models/meta-llama/Llama-3.3-70B-Instruct.toml
index 4644871f3d..f9838c63e5 100644
--- a/providers/nebius/models/meta-llama/Llama-3.3-70B-Instruct.toml
+++ b/providers/nebius/models/meta-llama/Llama-3.3-70B-Instruct.toml
@@ -1,25 +1,14 @@
-name = "Llama-3.3-70B-Instruct"
+base_model = "meta/llama-3.3-70b-instruct"
 attachment = false
-reasoning = false
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-08"
-release_date = "2025-12-05"
-last_updated = "2026-02-04"
-open_weights = true
+structured_output = false
 
 [cost]
 input = 0.13
-output = 0.40
+output = 0.4
 cache_read = 0.013
 cache_write = 0.16
 
 [limit]
-context = 128_000
+context = 131_072
 input = 120_000
 output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
\ No newline at end of file
diff --git a/providers/nebius/models/moonshotai/Kimi-K2.5-fast.toml b/providers/nebius/models/moonshotai/Kimi-K2.5-fast.toml
deleted file mode 100644
index 83dd3e8848..0000000000
--- a/providers/nebius/models/moonshotai/Kimi-K2.5-fast.toml
+++ /dev/null
@@ -1,31 +0,0 @@
-name = "Kimi-K2.5-fast"
-family = "kimi-k2"
-release_date = "2025-12-15"
-last_updated = "2026-02-04"
-attachment = true
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-06"
-open_weights = true
-
-[cost]
-input = 0.50
-output = 2.50
-cache_read = 0.05
-cache_write = 0.625
-
-[limit]
-context = 256_000
-input = 256_000
-output = 8192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/moonshotai/Kimi-K2.5.toml b/providers/nebius/models/moonshotai/Kimi-K2.5.toml
deleted file mode 100644
index b274d3a0b1..0000000000
--- a/providers/nebius/models/moonshotai/Kimi-K2.5.toml
+++ /dev/null
@@ -1,32 +0,0 @@
-name = "Kimi-K2.5"
-family = "kimi-k2"
-release_date = "2025-12-15"
-last_updated = "2026-02-04"
-attachment = true
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-06"
-open_weights = true
-
-[cost]
-input = 0.50
-output = 2.50
-reasoning = 2.50
-cache_read = 0.05
-cache_write = 0.625
-
-[limit]
-context = 256_000
-input = 256_000
-output = 8192
-
-[modalities]
-input = ["text", "image"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/moonshotai/Kimi-K2.6.toml b/providers/nebius/models/moonshotai/Kimi-K2.6.toml
new file mode 100644
index 0000000000..d6a08cfc6f
--- /dev/null
+++ b/providers/nebius/models/moonshotai/Kimi-K2.6.toml
@@ -0,0 +1,12 @@
+base_model = "moonshotai/kimi-k2.6"
+reasoning_options = []
+
+[cost]
+input = 0.95
+output = 4
+
+[limit]
+context = 8_000
+
+[modalities]
+input = ["text", "image"]
diff --git a/providers/nebius/models/moonshotai/Kimi-K2.7-Code.toml b/providers/nebius/models/moonshotai/Kimi-K2.7-Code.toml
new file mode 100644
index 0000000000..49cacd2cba
--- /dev/null
+++ b/providers/nebius/models/moonshotai/Kimi-K2.7-Code.toml
@@ -0,0 +1,11 @@
+base_model = "moonshotai/kimi-k2.7-code"
+attachment = false
+temperature = true
+reasoning_options = []
+
+[cost]
+input = 0.95
+output = 4
+
+[modalities]
+input = ["text"]
diff --git a/providers/nebius/models/nvidia/Cosmos3-Super-Reasoner.toml b/providers/nebius/models/nvidia/Cosmos3-Super-Reasoner.toml
new file mode 100644
index 0000000000..71c31012eb
--- /dev/null
+++ b/providers/nebius/models/nvidia/Cosmos3-Super-Reasoner.toml
@@ -0,0 +1,22 @@
+name = "Cosmos3-Super-Reasoner"
+release_date = "2026-07-01"
+last_updated = "2026-07-01"
+attachment = true
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+reasoning_options = []
+
+[cost]
+input = 0.1
+output = 0.3
+
+[limit]
+context = 8_000
+output = 8_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
diff --git a/providers/nebius/models/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1.toml b/providers/nebius/models/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1.toml
index 47b8177e2e..f6bab751b4 100644
--- a/providers/nebius/models/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1.toml
+++ b/providers/nebius/models/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1.toml
@@ -1,26 +1,15 @@
-name = "Llama-3.1-Nemotron-Ultra-253B-v1"
 base_model = "nvidia/llama-3.1-nemotron-ultra-253b"
-attachment = false
-reasoning = false
-tool_call = true
+tool_call = false
 structured_output = true
-temperature = true
-knowledge = "2024-12"
-release_date = "2025-01-15"
-last_updated = "2026-02-04"
-open_weights = true
+reasoning_options = []
 
 [cost]
-input = 0.60
-output = 1.80
+input = 0.6
+output = 1.8
 cache_read = 0.06
 cache_write = 0.75
 
 [limit]
-context = 128_000
+context = 8_000
 input = 120_000
 output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
\ No newline at end of file
diff --git a/providers/nebius/models/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B.toml b/providers/nebius/models/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B.toml
index f24f5276fe..d27efb0668 100644
--- a/providers/nebius/models/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B.toml
+++ b/providers/nebius/models/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B.toml
@@ -1,14 +1,6 @@
-name = "Nemotron-3-Nano-30B-A3B"
 base_model = "nvidia/nemotron-3-nano-30b-a3b"
-attachment = false
-reasoning = false
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-05"
-release_date = "2025-08-10"
-last_updated = "2026-02-04"
-open_weights = true
+structured_output = false
+reasoning_options = []
 
 [cost]
 input = 0.06
@@ -17,10 +9,5 @@ cache_read = 0.006
 cache_write = 0.075
 
 [limit]
-context = 32_000
 input = 30_000
 output = 4_096
-
-[modalities]
-input = ["text"]
-output = ["text"]
\ No newline at end of file
diff --git a/providers/nebius/models/nvidia/Nemotron-3-Nano-Omni.toml b/providers/nebius/models/nvidia/Nemotron-3-Nano-Omni.toml
index 114b15c208..b577986127 100644
--- a/providers/nebius/models/nvidia/Nemotron-3-Nano-Omni.toml
+++ b/providers/nebius/models/nvidia/Nemotron-3-Nano-Omni.toml
@@ -1,15 +1,7 @@
-name = "Nemotron-3-Nano-Omni"
 base_model = "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning"
 attachment = false
-reasoning = true
-reasoning_options = []
-tool_call = true
 structured_output = true
-temperature = true
-knowledge = "2025-01"
-release_date = "2025-01-20"
-last_updated = "2026-05-07"
-open_weights = true
+reasoning_options = []
 
 [cost]
 input = 0.06
@@ -18,10 +10,9 @@ cache_read = 0.006
 cache_write = 0.075
 
 [limit]
-context = 65_536
+context = 8_000
 input = 60_000
 output = 8_192
 
 [modalities]
 input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/nvidia/Nemotron-3-Ultra-550b-a55b.toml b/providers/nebius/models/nvidia/Nemotron-3-Ultra-550b-a55b.toml
new file mode 100644
index 0000000000..d06f8a9cbc
--- /dev/null
+++ b/providers/nebius/models/nvidia/Nemotron-3-Ultra-550b-a55b.toml
@@ -0,0 +1,10 @@
+base_model = "nvidia/nemotron-3-ultra-550b-a55b"
+structured_output = true
+reasoning_options = []
+
+[cost]
+input = 1
+output = 3
+
+[limit]
+context = 8_000
diff --git a/providers/nebius/models/nvidia/nemotron-3-super-120b-a12b.toml b/providers/nebius/models/nvidia/nemotron-3-super-120b-a12b.toml
index ae14965ad6..f93bc83302 100644
--- a/providers/nebius/models/nvidia/nemotron-3-super-120b-a12b.toml
+++ b/providers/nebius/models/nvidia/nemotron-3-super-120b-a12b.toml
@@ -1,25 +1,11 @@
-name = "Nemotron-3-Super-120B-A12B"
 base_model = "nvidia/nemotron-3-super-120b-a12b"
-attachment = false
-reasoning = true
-reasoning_options = []
-tool_call = true
 structured_output = true
-temperature = true
-knowledge = "2026-02"
-release_date = "2026-03-11"
-last_updated = "2026-03-12"
-open_weights = true
+reasoning_options = []
 
 [cost]
-input = 0.30
-output = 0.90
+input = 0.3
+output = 0.9
 
 [limit]
-context = 256_000
 input = 256_000
 output = 32_768
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/openai/gpt-oss-120b-fast.toml b/providers/nebius/models/openai/gpt-oss-120b-fast.toml
deleted file mode 100644
index 5cebac08cc..0000000000
--- a/providers/nebius/models/openai/gpt-oss-120b-fast.toml
+++ /dev/null
@@ -1,27 +0,0 @@
-name = "gpt-oss-120b-fast"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-06"
-release_date = "2025-06-10"
-last_updated = "2026-05-07"
-open_weights = true
-
-[cost]
-input = 0.10
-output = 0.50
-cache_read = 0.01
-cache_write = 0.125
-
-[limit]
-context = 8_000
-input = 7_000
-output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/nebius/models/openai/gpt-oss-120b.toml b/providers/nebius/models/openai/gpt-oss-120b.toml
index c73b1027e3..76f50c79ac 100644
--- a/providers/nebius/models/openai/gpt-oss-120b.toml
+++ b/providers/nebius/models/openai/gpt-oss-120b.toml
@@ -1,30 +1,19 @@
-name = "gpt-oss-120b"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2025-09"
-release_date = "2026-01-10"
-last_updated = "2026-02-04"
-open_weights = true
+base_model = "openai/gpt-oss-120b"
+
+[interleaved]
+field = "reasoning_content"
+
+[[reasoning_options]]
+type = "effort"
+values = ["low", "medium", "high"]
 
 [cost]
 input = 0.15
-output = 0.60
-reasoning = 0.60
+output = 0.6
+reasoning = 0.6
 cache_read = 0.015
 cache_write = 0.18
 
 [limit]
-context = 128_000
 input = 124_000
 output = 8_192
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/providers/nebius/models/openbmb/MiniCPM-V-4_5.toml b/providers/nebius/models/openbmb/MiniCPM-V-4_5.toml
new file mode 100644
index 0000000000..1932885092
--- /dev/null
+++ b/providers/nebius/models/openbmb/MiniCPM-V-4_5.toml
@@ -0,0 +1,21 @@
+name = "MiniCPM-V-4_5"
+release_date = "2026-07-01"
+last_updated = "2026-07-01"
+attachment = true
+reasoning = false
+temperature = true
+tool_call = false
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.658
+output = 1.11
+
+[limit]
+context = 8_000
+output = 8_000
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
diff --git a/providers/nebius/models/zai-org/GLM-5.1.toml b/providers/nebius/models/zai-org/GLM-5.1.toml
new file mode 100644
index 0000000000..978b632cc4
--- /dev/null
+++ b/providers/nebius/models/zai-org/GLM-5.1.toml
@@ -0,0 +1,9 @@
+base_model = "zhipuai/glm-5.1"
+reasoning_options = []
+
+[cost]
+input = 1.4
+output = 4.4
+
+[limit]
+context = 202_752
diff --git a/providers/nebius/models/zai-org/GLM-5.2.toml b/providers/nebius/models/zai-org/GLM-5.2.toml
index d50ba73852..0858ea7e39 100644
--- a/providers/nebius/models/zai-org/GLM-5.2.toml
+++ b/providers/nebius/models/zai-org/GLM-5.2.toml
@@ -1,16 +1,16 @@
 base_model = "zhipuai/glm-5.2"
 
+[interleaved]
+field = "reasoning_content"
+
 [[reasoning_options]]
 type = "effort"
 values = ["low", "medium", "high"]
 
-[interleaved]
-field = "reasoning_content"
-
 [cost]
 input = 1.4
 output = 4.4
 
 [limit]
-context = 432_000
+context = 1_048_576
 output = 432_000
diff --git a/providers/nebius/models/zai-org/GLM-5.toml b/providers/nebius/models/zai-org/GLM-5.toml
deleted file mode 100644
index 099a4056c0..0000000000
--- a/providers/nebius/models/zai-org/GLM-5.toml
+++ /dev/null
@@ -1,30 +0,0 @@
-name = "GLM-5"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-status = "deprecated"
-tool_call = true
-structured_output = true
-temperature = true
-knowledge = "2026-01"
-release_date = "2026-03-01"
-last_updated = "2026-03-10"
-open_weights = false
-
-[cost]
-input = 1.00
-output = 3.20
-cache_read = 0.10
-cache_write = 1.00
-
-[limit]
-context = 200_000
-input = 200_000
-output = 16_384
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"
diff --git a/sync.md b/sync.md
index c0e92693b0..db1d8ee119 100644
--- a/sync.md
+++ b/sync.md
@@ -93,7 +93,7 @@ Prefer small, provider-specific PRs when adding a provider. If the provider has
 
 ## Automation
 
-`.github/workflows/sync-models.yml` runs on an hourly schedule and manually through `workflow_dispatch`.
+`.github/workflows/sync-models.yml` runs on a daily schedule and manually through `workflow_dispatch`.
 
 The workflow:
 
@@ -107,7 +107,7 @@ The workflow:
 
 Each provider job checks out `dev` and writes to a fixed provider branch like `automation/sync-models-openrouter`. If that provider's sync PR is already open, later scheduled runs force-update the same branch and edit the existing PR instead of creating another one. Provider jobs do not share unmerged changes with each other; OpenRouter only uses `base_model` for model metadata entries already present on `dev`.
 
-CI automatically picks up providers registered in `providers` in `packages/core/src/sync/index.ts`. Adding a new sync provider there is enough to get an hourly provider-specific sync job, branch, labels, title, and PR naming convention. The workflow only needs manual updates when a new provider requires new secrets or other environment variables.
+CI automatically picks up providers registered in `providers` in `packages/core/src/sync/index.ts`. Adding a new sync provider there is enough to get a daily provider-specific sync job, branch, labels, title, and PR naming convention. The workflow only needs manual updates when a new provider requires new secrets or other environment variables.
 
 Actions are pinned by commit SHA. Keep new workflow actions pinned the same way.
 
@@ -155,6 +155,18 @@ xAI is implemented in `packages/core/src/sync/providers/xai.ts`.
 - Existing xAI models are updated from API-authoritative fields while local metadata is preserved for fields the API does not expose, especially output token limits and some feature/capability flags.
 - New xAI API models are reported in `.sync/model-sync-report.md` but not created automatically because the API does not provide enough authoritative metadata for complete catalog entries.
 
+## Nebius Token Factory Notes
+
+Nebius Token Factory is implemented in `packages/core/src/sync/providers/nebius.ts`.
+
+- Source endpoint: `https://api.tokenfactory.nebius.com/v1/models?verbose=true`; the `verbose=true` query parameter is required, since the default response only returns bare model IDs with no pricing, context length, or capability metadata.
+- Required auth: `NEBIUS_API_KEY`.
+- Model IDs map directly to TOML paths under `providers/nebius/models` (e.g. `openai/gpt-oss-120b` → `providers/nebius/models/openai/gpt-oss-120b.toml`).
+- The API is authoritative for `name` (a small `NEBIUS_ORG_TO_MODEL_PROVIDER`-independent fix strips org prefixes Nebius sometimes bakes into `name`, e.g. `openbmb/MiniCPM-V-4_5`), `cost` (converted from per-token to per-1M-token pricing), `limit.context`, `reasoning`/`tool_call`/`structured_output` (from `supported_features`), `temperature` (from `supported_sampling_parameters`, falling back to the existing flag when the list is empty), and `modalities`/`attachment` (parsed from `architecture.modality`, e.g. `text+image->text`; the `embedding` modality is mapped to `text` since the catalog schema has no embedding modality).
+- New remote models are created automatically. A small `NEBIUS_ORG_TO_MODEL_PROVIDER` map (scoped to orgs Nebius currently hosts) resolves known orgs (`meta-llama`, `Qwen`, `google`, `openai`, `nvidia`, `zai-org`, `moonshotai`, `MiniMaxAI`, `deepseek-ai`) to their `models/` namespace, and `factorBaseModel` (shared with OpenRouter) writes a `base_model` reference when a matching canonical entry exists — this is how new models get an authoritative output token limit despite the Models API not exposing one. New models with no canonical `models/` match (unmapped orgs, or a checkpoint not yet in `models/`) are still created as full entries, with `limit.output` falling back to the served context length.
+- `reasoning_options`, `limit.input`, `limit.output` (for already-factored/full existing entries), `knowledge`, `family`, `status`, `interleaved`, and cache pricing ratios are preserved from existing TOMLs since the API does not expose them.
+- `base_model`/`base_model_omit` are preserved from existing TOMLs, and newly resolved for brand-new models per the point above.
+
 ## OVHcloud Notes
 
 OVHcloud AI Endpoints is implemented in `packages/core/src/sync/providers/ovhcloud.ts`.