diff --git a/.changeset/astra-fable51.md b/.changeset/astra-fable51.md
new file mode 100644
index 00000000..021db981
--- /dev/null
+++ b/.changeset/astra-fable51.md
@@ -0,0 +1,10 @@
+---
+"@transloadit/node": minor
+"transloadit": minor
+"@transloadit/utils": patch
+"@transloadit/mcp-server": patch
+"@transloadit/types": minor
+"@transloadit/zod": minor
+---
+
+Support GPT-6 Astra and Claude Fable 5.1, and use Astra for automatic chat and assembly-instruction compilation. Assembly compilation retains medium reasoning effort. Existing model identifiers remain supported.
diff --git a/docs/prompts/2026-09-06-astra-fable51.md b/docs/prompts/2026-09-06-astra-fable51.md
new file mode 100644
index 00000000..61f0a287
--- /dev/null
+++ b/docs/prompts/2026-09-06-astra-fable51.md
@@ -0,0 +1,12 @@
+# Astra and Fable 5.1 rollout review
+
+PR: https://github.com/transloadit/node-sdk/pull/490
+
+- [x] Review PR context; no open human review comments.
+- [x] Council: add the missing release changeset, including companion packages.
+- [x] Council: replace the stale Sol-default comment.
+- [x] Council: record the backend dependency and block client rollout until API support is deployed.
+- [x] `yarn check` passed; Node 20/22/24, E2E, build, and release dry-run CI passed on the implementation head.
+
+This PR must not release new client defaults before the API2 sister PR is deployed. Current CI and
+deployment/merge status remain on the PR.
diff --git a/packages/node/src/alphalib/types/robots/ai-chat.ts b/packages/node/src/alphalib/types/robots/ai-chat.ts
index 1bc01b20..a0802d38 100644
--- a/packages/node/src/alphalib/types/robots/ai-chat.ts
+++ b/packages/node/src/alphalib/types/robots/ai-chat.ts
@@ -551,6 +551,7 @@ export const MODEL_CAPABILITIES: Record<string, { pdf: boolean; image: boolean }
   'anthropic/claude-opus-4-6': { pdf: true, image: true },
   'anthropic/claude-opus-4-7': { pdf: true, image: true },
   'anthropic/claude-fable-5': { pdf: true, image: true },
+  'anthropic/claude-fable-5-1': { pdf: true, image: true },
   'anthropic/claude-sonnet-5': { pdf: true, image: true },
   'openai/gpt-4.1-2025-04-14': { pdf: false, image: true },
   'openai/chatgpt-4o-latest': { pdf: false, image: true },
@@ -564,6 +565,7 @@ export const MODEL_CAPABILITIES: Record<string, { pdf: boolean; image: boolean }
   'openai/gpt-5.2-pro': { pdf: false, image: true },
   'openai/gpt-5.5': { pdf: false, image: true },
   'openai/gpt-5.6-sol': { pdf: false, image: true },
+  'openai/gpt-6-astra': { pdf: false, image: true },
   'openai/gpt-5.4': { pdf: false, image: true },
   'openai/gpt-5.4-mini': { pdf: false, image: true },
   'openai/gpt-5.4-nano': { pdf: false, image: true },
@@ -571,10 +573,8 @@ export const MODEL_CAPABILITIES: Record<string, { pdf: boolean; image: boolean }
   'moonshot/kimi-k2': { pdf: false, image: false },
 }
 
-// Default model for /ai/chat when `model: "auto"` (or unset).
-// 2026-07-09: default is GPT-5.6 Sol (intentional; aligns with our current recommended OpenAI
-// flagship model). Keep this aligned with MODEL_CAPABILITIES.
-export const AI_CHAT_DEFAULT_MODEL = 'openai/gpt-5.6-sol' satisfies keyof typeof MODEL_CAPABILITIES
+/** Default /ai/chat model for `auto` or omitted selections; keep aligned with the API backend. */
+export const AI_CHAT_DEFAULT_MODEL = 'openai/gpt-6-astra' satisfies keyof typeof MODEL_CAPABILITIES
 
 const supportedModelsList = Object.keys(MODEL_CAPABILITIES)
 
@@ -614,7 +614,7 @@ export const robotAiChatInstructionsSchema = robotBase
       .enum(['xhigh', 'high', 'medium', 'low'])
       .optional()
       .describe(
-        'Controls how much effort the model spends on reasoning. Higher values produce more thorough responses but cost more tokens. Applies to models that support extended thinking (OpenAI o-series, GPT-5.x, Anthropic Claude with thinking). If omitted, the model default is used.',
+        'Controls how much effort the model spends on reasoning. Higher values produce more thorough responses but cost more tokens. Applies to models that support extended thinking (OpenAI o-series, GPT-5.x, GPT-6, Anthropic Claude with thinking). If omitted, the model default is used.',
       ),
     credentials: z
       .union([z.string(), z.array(z.string())])
diff --git a/packages/node/test/unit/assembly-instructions-compiler.test.ts b/packages/node/test/unit/assembly-instructions-compiler.test.ts
index bf002b98..c32f6a83 100644
--- a/packages/node/test/unit/assembly-instructions-compiler.test.ts
+++ b/packages/node/test/unit/assembly-instructions-compiler.test.ts
@@ -68,7 +68,7 @@ describe('compileAssemblyInstructionsFromPrompt', () => {
     expect(aiSteps[0]).toMatchObject({
       robot: '/ai/chat',
       result: true,
-      model: 'openai/gpt-5.6-sol',
+      model: 'openai/gpt-6-astra',
       reasoning_effort: 'medium',
       format: 'json',
       interpolate: { system_message: false },
diff --git a/packages/utils/src/assemblyInstructionsCompiler.ts b/packages/utils/src/assemblyInstructionsCompiler.ts
index 084f0c66..49cb9bc6 100644
--- a/packages/utils/src/assemblyInstructionsCompiler.ts
+++ b/packages/utils/src/assemblyInstructionsCompiler.ts
@@ -1,4 +1,4 @@
-export const COMPILE_ASSEMBLY_INSTRUCTIONS_DEFAULT_MODEL = 'openai/gpt-5.6-sol'
+export const COMPILE_ASSEMBLY_INSTRUCTIONS_DEFAULT_MODEL = 'openai/gpt-6-astra'
 export const COMPILE_ASSEMBLY_INSTRUCTIONS_DEFAULT_REASONING_EFFORT = 'medium'
 export const COMPILE_ASSEMBLY_INSTRUCTIONS_DEFAULT_MAX_ATTEMPTS = 3