R
RADIXARK / RadixArk/DaisyRose-27B

DaisyRose 27B

RadixArk multimodal model introduced on November 18. It starts from Qwen3.8-27B and uses reinforcement learning on long agent tasks. It retains image understanding and 256K context; 1M training sessions do not enlarge that limit.

01 / SUMMARY

The essential

ORIENTATIONQwen3.8 reinforcement post-training

Primary use stated or inferred cautiously from official documentation.

CONTEXT262,144 tokens

Maximum input capacity when published by the source.

EXITText; separate output maximum unpublished

Documented maximum generation limit.

ACCESSHugging Face · downloadable checkpoint

Verified availability channels.

MODEL TYPEMultimodal · Code · Text and reasoning

Primary functional family and verified specialties.

LICENSEApache 2.0

Declared terms for API access, model weights, or self-hosting.

02 / TECHNICAL SHEET

Limits and integration

API IDRadixArk/DaisyRose-27B
Model typeMultimodal · Code · Text and reasoning
Access modelOpen weights
LicenseApache 2.0
DeploymentSelf-hosted
Release18 NOV 2026
Knowledge cutoffNot published
EntranceText · Image
ExitText
Context window262,144 tokens
maximum outputText; separate output maximum unpublished
ReasoningNot published
Published tools
Structured OutingsNot published
Batch processingNot published
Prompt cacheNot published
Fine-tuningNot published
Verified platformsHugging Face
BEST SUITED FOR

Coding, tool use and adaptation to agent environments.

WORTH MONITORING

The complete architecture includes vision and multi-token prediction: about 27.8B total parameters. BF16, FP8 and NVFP4 are formats of the same model. Long training tasks do not guarantee reliability in unlimited autonomous sessions.

03 / CAPABILITIES

What it can do

01

Orientation

Coding, tool use and adaptation to agent environments.

02

Context

262,144 tokens · Text; separate output maximum unpublished

03

Tools and integration

Not published

04

Access

Hugging Face · downloadable checkpoint

Input modalities
TextImage
04 / PRICES

Documented cost

ConceptWorthUnit/condition
Standard inputNo published official API rateStandard API rate
Cached inputNot publishedReading reused prefixes
Cache write or storageNot publishedThe condition varies by provider
Standard outputSelf-hosted infrastructure costMay include reasoning tokens
Batch inputNot publishedAsynchronous processing
Batch outputNot publishedAsynchronous processing

No verified official API tariff. Self-hosting requires memory, compute and maintenance; downloadable weights do not make inference free.

i Prices change and may depend on level, region or context length. Check the source before making a decision.

05 / EVALUATIONS

How to read the results

“

RadixArk reports 46.2% on DeepSWE v1.1 with eight repetitions per task and mini-swe-agent, and 90.7% on GPQA Diamond with eight repetitions of 198 questions. The table takes the better of BF16 and FP8; it is not a single-precision result or an Inferama evaluation.

BenchmarkResultMetricSource
DeepSWE v1.146.2 %Published scoreView source ↗
GPQA Diamond90.7 %Published scoreView source ↗

Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.

06 / VERSIONS

Chronology

radixark/DaisyRose-27B

Version added to Inferama's verified catalog.

07 / SOURCES

Traceability