Bonus: Top up now and we'll double your first deposit — get x2 credits instantly.
Get Started
Sign Up
Sign Up

Cydonia 24B V4.1: Self-Hosting & Deployment Guide

Cydonia 24B V4.1 is a 24-billion-parameter, open-source text language model by TheDrummer, fine-tuned from Mistral Small 3.2 and optimized for uncensored creative writing with a 131K-token context window.

Browse available models

What is Cydonia 24B V4.1?

Cydonia 24B V4.1 is an open‑source, text‑to‑text language model by TheDrummer built on Mistral Small 3.2 24B with a ~131K token context window. It is primarily used for uncensored creative writing, roleplay, and narrative-heavy chat where mood, nuance, and consistent characterization matter over long conversations. It is also applied as a general-purpose assistant model in enterprise and hobbyist settings, offering relatively low per-token costs for large-context workloads. Cydonia 24B V4.1 continues TheDrummer’s Cydonia series, improving on earlier variants such as Cydonia-22B and Cydonia-24B-v2.x in focus, coherence, and writing quality.

How to run Cydonia 24B V4.1 yourself

Open weights, 23.6B parameters. Below: how much memory the weights need at each quantization, which hardware fits them, and the commands to serve it.

Cydonia 24B V4.1 VRAM requirements

Precision / quantizationWeightsFigureSource
BF16 (full precision)47.1 GBEstimate23.6B params × 2 bytes
FP824.9 GBFile sizeTheClusterDev/Cydonia-24B-v4.1-FP8
MLX 4-bit13.3 GBFile sizegg000/Cydonia-24B-v4.1-MLX-4bit
GGUF Q8_025.1 GBFile sizebartowski/TheDrummer_Cydonia-24B-v4.1-GGUF
GGUF Q4_K_M14.3 GBFile sizebartowski/TheDrummer_Cydonia-24B-v4.1-GGUF
GGUF Q2_K8.9 GBFile sizebartowski/TheDrummer_Cydonia-24B-v4.1-GGUF

File size = actual download size on Hugging Face. Estimate = parameter count × bytes per weight. Weights only — the KV cache for long contexts needs extra memory; the fit tables below add 15% headroom as a rule of thumb.

GPU requirements: deploy Cydonia 24B V4.1 on your hardware

HardwareBF16 (full precision)FP8MLX 4-bitGGUF Q8_0GGUF Q4_K_MGGUF Q2_K
NVIDIA H200141 GB1×1×1×1×1×1×
NVIDIA H10080 GB1×1×1×1×1×1×
NVIDIA A100 80GB80 GB1×1×1×1×1×1×
NVIDIA L40S48 GB2×1×1×1×1×1×

GPUs needed per variant.

HardwareGGUF Q8_0GGUF Q4_K_MGGUF Q2_K
RTX 509032 GB1×1×1×
RTX 409024 GB2×1×1×
RTX 309024 GB2×1×1×

GGUF via llama.cpp; more cards or CPU offload for larger files.

HardwareGGUF Q8_0GGUF Q4_K_MGGUF Q2_KMLX 4-bit
Mac, 512 GBM3 UltraFitsFitsFitsFits
Mac, 192 GBM2 UltraFitsFitsFitsFits
Mac, 128 GBM4 Max / M3 MaxFitsFitsFitsFits
Mac, 64 GBM4 Pro / MaxFitsFitsFitsFits

Assumes ~75% of unified memory is usable by the GPU.

HardwareBF16 (full precision)FP8MLX 4-bitGGUF Q8_0GGUF Q4_K_MGGUF Q2_K
AWS p5.48xlarge8× H100 · 640 GBFitsFitsFitsFitsFitsFits
AWS p4de.24xlarge8× A100 80GB · 640 GBFitsFitsFitsFitsFitsFits
AWS g6e.12xlarge4× L40S · 192 GBFitsFitsFitsFitsFitsFits
Google Cloud a3-highgpu-8g8× H100 · 640 GBFitsFitsFitsFitsFitsFits
Azure ND H100 v58× H100 · 640 GBFitsFitsFitsFitsFitsFits

Whole-instance GPU memory; public instance specs.

Not on LLM.API yet

We do not route this model through our API at the moment. Browse the models you can call today — most workloads have a close match already live.

Browse available models

Providers

Different companies host the same model. OpenRouter routes your request to one of them based on the routing mode you pick — Balanced (price + speed), Nitro (fastest), or Exacto (highest tool-calling accuracy).

List price by provider

InputOutput
Parasail$0.0003 in
$0.0005 out

Provider list prices; the LLM.API discount applies on top.

Provider Input Output Cache read /M Latency Throughput Uptime
Parasailin $0.0003; out $0.0005 per 1M tokens131K tokensstreaming, JSON, structured

List prices from third-party hosts in the OpenRouter public catalogue (thedrummer/cydonia-24b-v4.1); LLM API does not serve this model and applies no discount. Updated weekly. Last updated 11 Oct 2026.

Try this model

Test Cydonia 24B V4.1 right here — free to start.

Cydonia 24B V4.1
Hi! Want to test the model?

Suggestions for your first prompt

Not available on LLM API yet

We do not route this model through our API at the moment, so there is no endpoint or code snippet for it yet. Browse the models you can call today — most workloads have a close match already live.

Browse available models

Cydonia 24B V4.1: Self-Hosting & Deployment Guide in production

What the model is good at.

  • Conversational Chat

    Engages in multi-turn dialogue, answering questions, following instructions, and maintaining context within general conversational and assistant tasks.

  • Code and Logs

    Reads and writes code or technical text, explaining behavior, debugging issues, and providing structured suggestions within its training scope.

  • Visual Content

    Processes image inputs to identify objects and scenes and provide descriptive text responses within its supported visual understanding abilities.

  • Optical Text Reading

    Extracts readable text from images or screenshots and converts it into machine-readable form for further processing or analysis.

  • Language Translation

    Translates written text between multiple languages, preserving meaning and tone as closely as possible within its training limitations.

What you get when you call Cydonia 24B V4.1: Self-Hosting & Deployment Guide through LLM.API.

  • Intelligent AI Routing

    Dynamically route each request across providers and models based on latency, cost, and quality—without changing your integration.

    One endpoint, any model

  • Cost-Aware Orchestration

    Automatically pick the most cost-effective model for each task and track spend per project, environment, and feature in one place.

    Optimize tokens, not code

  • Resilient Fallback Flows

    Define fallback chains across providers so requests transparently recover from outages, rate limits, and model regressions.

    Keep responses flowing

  • Full-Stack Observability

    Get end-to-end traces, latency and error metrics, and model-level analytics to debug prompts and production traffic in real time.

    See every token hop

  • Task-Level Abstractions

    Describe tasks like chat, tool use, search, or generation once, then plug in any model or provider behind the same interface.

    Ship tasks, not glue code

  • High-Throughput Batch APIs

    Fan out thousands of requests per call with built-in retries, rate management, and structured result aggregation.

    Scale from 10 to 10M calls

When to Use — When NOT to Use

Use it if...

  • You need a mid-sized 24B model balancing capability with more moderate hardware requirements.
  • You need a community-driven open model that can be self-hosted and customized.
  • Your use case involves general coding assistance, debugging, and small-to-medium code generation.
  • Your use case involves chat-style assistants for customer support or internal knowledge bases.
  • You need an experimentation model for fine-tuning or benchmarking against other 20–30B models.
  • Your use case involves educational tutoring, explanations, and walkthroughs of technical concepts.

Avoid if...

  • You need frontier-level reasoning comparable to the very latest large proprietary flagship models.
  • Your workload requires extremely long context handling, such as full-book ingestion or analysis.
  • You need highly specialized domain performance in law, medicine, or finance with certifications.
  • Your workload requires ultra-low latency inference at massive scale on very limited hardware.
  • You need guaranteed first-party support, SLAs, and an enterprise governance or compliance program.
  • Your workload requires cutting-edge multimodal capabilities like advanced vision, audio, or video understanding.

COMMUNITY

What developers say about Cydonia 24B V4.1

LMArena—Speed—Sources: LMArena, Artificial Analysis

Summarised from public developer discussions and write-ups. Opinions are the sources’, not LLM.API’s.

  • Developers on Hugging Face describe the model as a significant 'level up' for creative writing, praising its ability to capture mood and nuance without resorting to overly complex 'purple prose.'
  • Technical users note that the model avoids common 'Mistral tuneisms' and 'slop,' demonstrating strong focus and accurate roleplay capabilities even at 6k to 12k context lengths.
  • Builders emphasize that while the 131K context window is excellent for sustained narratives, a 1.60% error rate makes it less suitable for strict JSON-schema or high-stakes structured output pipelines.

SOURCES

Frequently Asked Questions

  • What is Cydonia 24B V4.1?

    Cydonia 24B V4.1 is a 24-billion-parameter language model by TheDrummer focused on fast, general-purpose code and text generation via LLM.API.

  • What is Cydonia 24B V4.1 best suited for?

    It is best for code completion, technical writing, tool-using agents, and structured data generation where latency and cost matter.

  • What is the context window of Cydonia 24B V4.1?

    Cydonia 24B V4.1 supports a context window of up to 32,000 tokens per request.

  • What modalities does Cydonia 24B V4.1 support?

    Cydonia 24B V4.1 is a text-only model that accepts and outputs UTF-8 text.

  • How is Cydonia 24B V4.1 priced on LLM.API?

    Pricing is usage-based per 1,000 tokens, with separate rates for input and output tokens defined in your LLM.API account.

  • How fast is Cydonia 24B V4.1 in production use?

    Typical end-to-end latency is in the low hundreds of milliseconds for short prompts, depending on load and request size.

  • How do I call Cydonia 24B V4.1 through LLM.API?

    Specify the model name "TheDrummer/cydonia-24b-v4.1" in your LLM.API completion or chat endpoint requests with your API key.

  • How does Cydonia 24B V4.1 compare to similar 20–30B models?

    It targets a balance of stronger coding ability and lower latency than many open 20–30B models at similar price points.

  • Does Cydonia 24B V4.1 support function calling or tools via LLM.API?

    Yes, you can use LLM.API's tool or function-calling conventions with this model for agent-style workflows.

  • What are the main limitations of Cydonia 24B V4.1?

    It may hallucinate facts, lacks real-time knowledge, and is not guaranteed safe for high-stakes decisions without human review.

COMPARE

Competitive Models

Get one key to every model

Swap your API key. Keep your code.

Browse available models Talk to sales