Full Guide: Re-create Fable 5 At Half The Cost

One of Fable’s biggest selling points was its excellent coding and autonomous work capabilities.

If you’ve been paying attention online, you’ll know that GLM-5.2 was just released, and it’s nearly as strong as Fable for coding-related tasks:

This new GLM model is built specifically for long-horizon, autonomous work like Fable.

To get a similar Fable 5 effect, you can inject this model into Claude Code while running /goal & /loop commands to get a similar “fully autonomous” Fable feel.

How to Run GLM-5.2 in Claude Code (step-by-step):

Step 1: Subscribe to zAI GLM Coding Plan

The GLM Coding Plan has three main tiers: Lite, Pro, and Max.

These plans measure usage in prompts per cycle, not tokens.

Step 2: Install Claude Code

If you don't have Claude Code installed yet, run this in your terminal:

Step 3: Get your zAI API Key

Go to: https://z.ai/model-api

Create a new API key and copy it.

Step 4: Configure environment

Run this in your terminal:

Step 5: Verify & Deploy

Enter the /status command to check the current model status and confirm GLM-5.2 is running.

Done! Now you can start using /goal & /loop prompts in Claude Code, with GLM-5.2 underneath, to replicate Fable-level coding.

With this setup, you’re running Fable 5 intelligence at ~50% of the cost.

Note: This model is primarily built for coding tasks - it is not a great conversational model like Claude/GPT. We only recommend running this setup for coding your work.

For easier setup, you can send this prompt to Claude Code terminal:

Method #1 is the best strategy for achieving Fable-level intelligence for coding tasks. However, it lacks general conversational skills.

This next method is best for re-creating Fable 5’s general conversational abilities (planning, research, etc.)

Last week, OpenRouter deployed its new “Fusion” API.

Fusion API is an orchestration of multiple models that sends a single prompt to multiple AI agents in parallel, analyzes their responses, and synthesizes them into one high-quality final answer.

Essentially, you’re combining LLMs into an agent swarm that is more well-rounded and intelligent.

When used in the latest frontier models, the result is Fable 5 intelligence at half the cost:

You have two options to set up Fusion API:

Go to: https://openrouter.ai/fusion

Sign in with your OpenRouter account (free to start, add credits as needed).

Click "add model" and add "Fusion" (at the top of the list).

Start chatting normally.

Prerequisites:

API setup (step-by-step)

Extremely simple to set up, and this is the closest thing you'll get to all-around Fable intelligence right now.

When Fable 5 went live, someone leaked the entire system prompt built by Anthropic.

To be clear, just because you have a system prompt doesn’t mean you have the model.

However, the system prompt is what instructs the LLM on how it should behave.

So, in theory, you can take the Fable 5 system prompt, inject it into other models, and have those models mimic responses as if they were Fable.

How to set up

Lastly, you want to paste the entire system prompt into any other LLM.

For example, you can send the prompt to Opus 4.8 on xHigh effort, and the model will start acting like Fable (with Opus 4.8's underlying intelligence).

This method is quite unique, and when paired with method #1 or #2, you can get extremely similar Fable 5 responses with excellent intelligence (and at a fraction of the cost).

It was a huge week in AI, so you won’t want to miss this!

OpenAI just announced its GPT-5.5 “Cyber” model.

Sam Altman: “Codex Security will help solve security problems instead of just finding them.”

Anthropic released Claude Tag in beta on June 23 for Claude Enterprise and Team customers.

Teams can tag @Claude inside Slack, give it access to selected channels, tools, and codebases, and let it complete multi-stage assignments before returning the result to the same thread.

Official announcement

Tokyo-based Sakana AI just released its Fugu model family.

Fugu acts as an agent coordinator that selects and combines several specialist models into a single OpenAI-compatible API (similar to the Fusion API).

The model stands toe-to-toe with Fable/Mythos class models (it’s also open-weight).

Try it here

Seedance 2.0 can now generate AI videos in 4K (up to 30-seconds).

A huge leap forward in video generation, even more powerful than Google Omni.

You can use the best AI video generation model here:

Anthropic appears to be prepping the next Sonnet model - expected as early as next week.

The full GPT 5.5 Cyber model is currently limited to trusted defenders.

It’s possible this new model will be live to the public in the coming weeks before GPT-5.6.

For the first time since the Fable ban, Trump has spoken out about the situation.

After watching this clip, we’re much more hopeful that Fable 5 will be back online soon.

Interview clip: