feat(claude): support [1m] suffix for 1M context window

Parse [1m] suffix from model name (e.g. claude-sonnet-4-6[1m]) and propagate extendedContext flag through the request pipeline to append context-1m-2025-08-07 to the Anthropic-Beta header.
2026-03-10 23:57:13 -03:00
parent c7ae9c30c2
commit 623c63baf6
5 changed files with 80 additions and 18 deletions
@@ -40,6 +40,7 @@ export type ExecuteInput = {
  credentials: ProviderCredentials;
  signal?: AbortSignal | null;
  log?: ExecutorLog | null;
+  extendedContext?: boolean;
 };

 function mergeAbortSignals(primary: AbortSignal, secondary: AbortSignal): AbortSignal {
@@ -174,7 +175,7 @@ export class BaseExecutor {
    return { status: response.status, message: bodyText || `HTTP ${response.status}` };
  }

-  async execute({ model, body, stream, credentials, signal, log }: ExecuteInput) {
+  async execute({ model, body, stream, credentials, signal, log, extendedContext }: ExecuteInput) {
    const fallbackCount = this.getFallbackCount();
    let lastError: unknown = null;
    let lastStatus = 0;
@@ -182,6 +183,29 @@ export class BaseExecutor {
    for (let urlIndex = 0; urlIndex < fallbackCount; urlIndex++) {
      const url = this.buildUrl(model, stream, urlIndex, credentials);
      const headers = this.buildHeaders(credentials, stream);
+
+      // Append 1M context beta header when [1m] suffix was used
+      // Only supported for specific Claude models per Anthropic docs
+      if (extendedContext) {
+        const EXTENDED_CONTEXT_MODELS = [
+          "claude-opus-4-6",
+          "claude-sonnet-4-6",
+          "claude-sonnet-4-5",
+          "claude-sonnet-4",
+        ];
+        const baseModel = model.replace(/-\d{8}$/, "");
+        if (
+          EXTENDED_CONTEXT_MODELS.some((m) => baseModel === m || model === m || model.startsWith(m))
+        ) {
+          const existing = headers["Anthropic-Beta"];
+          if (existing) {
+            headers["Anthropic-Beta"] = existing + ",context-1m-2025-08-07";
+          } else {
+            headers["Anthropic-Beta"] = "context-1m-2025-08-07";
+          }
+        }
+      }
+
      const transformedBody = this.transformRequest(model, body, stream, credentials);

      try {
@@ -68,7 +68,7 @@ export async function handleChatCore({
  userAgent,
  comboName,
 }) {
-  const { provider, model } = modelInfo;
+  const { provider, model, extendedContext } = modelInfo;
  const startTime = Date.now();

  // ── Phase 9.2: Idempotency check ──
@@ -270,6 +270,7 @@ export async function handleChatCore({
        credentials,
        signal: streamController.signal,
        log,
+        extendedContext,
      })
    );

@@ -357,6 +358,7 @@ export async function handleChatCore({
          credentials,
          signal: streamController.signal,
          log,
+          extendedContext,
        });

        if (retryResult.response.ok) {
@@ -59,29 +59,50 @@ function resolveProviderModelAlias(providerOrAlias, modelId) {

 /**
 * Parse model string: "alias/model" or "provider/model" or just alias
+ * Supports [1m] suffix for extended 1M context window (e.g. "claude-sonnet-4-6[1m]")
 */
 export function parseModel(modelStr) {
  if (!modelStr) {
-    return { provider: null, model: null, isAlias: false, providerAlias: null };
+    return {
+      provider: null,
+      model: null,
+      isAlias: false,
+      providerAlias: null,
+      extendedContext: false,
+    };
  }

  // Sanitize: reject strings with path traversal or control characters
  if (/\.\.[\/\\]/.test(modelStr) || /[\x00-\x1f]/.test(modelStr)) {
    console.log(`[MODEL] Warning: rejected malformed model string: "${modelStr.substring(0, 50)}"`);
-    return { provider: null, model: null, isAlias: false, providerAlias: null };
+    return {
+      provider: null,
+      model: null,
+      isAlias: false,
+      providerAlias: null,
+      extendedContext: false,
+    };
+  }
+
+  // Extract [1m] suffix before parsing provider/model
+  let extendedContext = false;
+  let cleanStr = modelStr;
+  if (cleanStr.endsWith("[1m]")) {
+    extendedContext = true;
+    cleanStr = cleanStr.slice(0, -4);
  }

  // Check if standard format: provider/model or alias/model
-  if (modelStr.includes("/")) {
-    const firstSlash = modelStr.indexOf("/");
-    const providerOrAlias = modelStr.slice(0, firstSlash);
-    const model = modelStr.slice(firstSlash + 1);
+  if (cleanStr.includes("/")) {
+    const firstSlash = cleanStr.indexOf("/");
+    const providerOrAlias = cleanStr.slice(0, firstSlash);
+    const model = cleanStr.slice(firstSlash + 1);
    const provider = resolveProviderAlias(providerOrAlias);
-    return { provider, model, isAlias: false, providerAlias: providerOrAlias };
+    return { provider, model, isAlias: false, providerAlias: providerOrAlias, extendedContext };
  }

  // Alias format (model alias, not provider alias)
-  return { provider: null, model: modelStr, isAlias: true, providerAlias: null };
+  return { provider: null, model: cleanStr, isAlias: true, providerAlias: null, extendedContext };
 }

 /**
@@ -123,12 +144,14 @@ export function resolveModelAliasFromMap(alias, aliases) {
 */
 export async function getModelInfoCore(modelStr, aliasesOrGetter) {
  const parsed = parseModel(modelStr);
+  const { extendedContext } = parsed;

  if (!parsed.isAlias) {
    const canonicalModel = resolveProviderModelAlias(parsed.provider, parsed.model);
    return {
      provider: parsed.provider,
      model: canonicalModel,
+      extendedContext,
    };
  }

@@ -142,6 +165,7 @@ export async function getModelInfoCore(modelStr, aliasesOrGetter) {
    return {
      provider: resolved.provider,
      model: canonicalModel,
+      extendedContext,
    };
  }

@@ -153,6 +177,7 @@ export async function getModelInfoCore(modelStr, aliasesOrGetter) {
    return {
      provider: "openai",
      model: modelId,
+      extendedContext,
    };
  }

@@ -160,7 +185,7 @@ export async function getModelInfoCore(modelStr, aliasesOrGetter) {
  if (nonOpenAIProviders.length === 1) {
    const provider = nonOpenAIProviders[0];
    const canonicalModel = resolveProviderModelAlias(provider, modelId);
-    return { provider, model: canonicalModel };
+    return { provider, model: canonicalModel, extendedContext };
  }

  if (nonOpenAIProviders.length > 1) {
@@ -182,5 +207,6 @@ export async function getModelInfoCore(modelStr, aliasesOrGetter) {
  return {
    provider: "openai",
    model: modelId,
+    extendedContext,
  };
 }
@@ -227,7 +227,7 @@ async function handleSingleModelChat(
  const resolved = await resolveModelOrError(modelStr, body);
  if (resolved.error) return resolved.error;

-  const { provider, model, sourceFormat, targetFormat } = resolved;
+  const { provider, model, sourceFormat, targetFormat, extendedContext } = resolved;

  // 2. Pipeline gates (availability + circuit breaker)
  const gate = checkPipelineGates(provider, model);
@@ -290,6 +290,7 @@ async function handleSingleModelChat(
      apiKeyInfo,
      userAgent,
      comboName,
+      extendedContext,
    });
    if (telemetry) telemetry.endPhase();

@@ -366,7 +367,7 @@ async function resolveModelOrError(modelStr: string, body: any) {
    return { error: errorResponse(HTTP_STATUS.BAD_REQUEST, "Invalid model format") };
  }

-  const { provider, model } = modelInfo;
+  const { provider, model, extendedContext } = modelInfo;
  const sourceFormat = detectFormat(body);
  const providerAlias = PROVIDER_ID_TO_ALIAS[provider] || provider;

@@ -378,13 +379,14 @@ async function resolveModelOrError(modelStr: string, body: any) {
    log.info("ROUTING", `Custom model apiFormat=responses → targetFormat=openai-responses`);
  }

+  const ctxTag = extendedContext && providerAlias === "claude" ? " [1m]" : "";
  if (modelStr !== `${provider}/${model}`) {
-    log.info("ROUTING", `${modelStr} → ${provider}/${model}`);
+    log.info("ROUTING", `${modelStr} → ${provider}/${model}${ctxTag}`);
  } else {
-    log.info("ROUTING", `Provider: ${provider}, Model: ${model}`);
+    log.info("ROUTING", `Provider: ${provider}, Model: ${model}${ctxTag}`);
  }

-  return { provider, model, sourceFormat, targetFormat };
+  return { provider, model, sourceFormat, targetFormat, extendedContext };
 }

 /**
@@ -437,6 +439,7 @@ async function executeChatWithBreaker({
  apiKeyInfo,
  userAgent,
  comboName,
+  extendedContext,
 }: any): Promise<{ result: any; tlsFingerprintUsed: boolean }> {
  let tlsFingerprintUsed = false;

@@ -445,7 +448,7 @@ async function executeChatWithBreaker({
      runWithProxyContext(proxyInfo?.proxy || null, () =>
        (handleChatCore as any)({
          body: { ...body, model: `${provider}/${model}` },
-          modelInfo: { provider, model },
+          modelInfo: { provider, model, extendedContext },
          credentials: refreshedCredentials,
          log: logger,
          clientRawRequest,
@@ -39,6 +39,7 @@ async function lookupCustomModelApiFormat(
 */
 export async function getModelInfo(modelStr) {
  const parsed = parseModel(modelStr);
+  const { extendedContext } = parsed;

  // Check custom provider nodes first (for both alias and non-alias formats)
  if (parsed.providerAlias || parsed.provider) {
@@ -53,7 +54,12 @@ export async function getModelInfo(modelStr) {
        matchedOpenAI.id as string,
        parsed.model as string
      );
-      return { provider: matchedOpenAI.id, model: parsed.model, ...(apiFormat && { apiFormat }) };
+      return {
+        provider: matchedOpenAI.id,
+        model: parsed.model,
+        extendedContext,
+        ...(apiFormat && { apiFormat }),
+      };
    }

    // Check Anthropic Compatible nodes
@@ -67,6 +73,7 @@ export async function getModelInfo(modelStr) {
      return {
        provider: matchedAnthropic.id,
        model: parsed.model,
+        extendedContext,
        ...(apiFormat && { apiFormat }),
      };
    }