Helios 1

Our most capable model. Helios 1 thinks before it answers, keeps the most of your chat in mind, and writes the longest, most detailed replies. It's built on a 27-billion-parameter open-weight base with its refusal behaviour removed through abliteration.

Helios 1, our frontier model
Parameters
27B
Context
32k tokens
Longest reply
~2,500 words
Memory
80 messages
Thinking
Visible
Plans
Max, Max 20x

Abliterated: the refusal direction is removed from the weights.

What it can do.

Reasoning

Works through the problem first. Open the thinking to see how it got there.

Long-form writing

Chapters, scripts and long answers, up to ~2,500 words in one reply.

Instruction following

Keeps to the format, tone and rules you set, even across a long chat.

Coding

Writes, explains and fixes code, from snippets to whole functions.

How it's set up.

Helios 1 next to our other models, as each one runs on Bleepless.

Longest reply (words)

Zouk 1~600
Zouk 2~1,200
Helios 1~2,500
Words per reply at most. Shorter questions get shorter answers.

Chat memory (messages)

Zouk 116
Zouk 240
Helios 180
How many recent messages the model re-reads on every reply.

Context window (tokens)

Zouk 124k
Zouk 224k
Helios 132k
The most text the model can hold at once, in tokens (about ¾ of a word each).

Benchmarks.

Base-model benchmarks (developer-reported)

  • GPQA DiamondGraduate-level science questions89.2
  • LiveCodeBench v6Competitive coding90.3
  • IFBenchInstruction following79.5
  • HLEVery hard, multidisciplinary reasoning30.8
Measured by the base model's developer. Helios 1 is abliterated and runs in 8-bit, which can shift scores by a point or two.

Strengths

  • Thinks step by step before answering, and you can read its thinking
  • The longest replies on Bleepless, up to ~2,500 words
  • The deepest memory: the last 80 messages and a 32k-token context
  • Strong on science, coding and following detailed instructions

Best for

  • Long-form stories and roleplay that need to stay consistent
  • Hard, detailed questions
  • Anything where depth matters more than speed

Questions about Helios 1.

What is Helios 1?

Helios 1 is the frontier model from Bleepless AI. It's a 27-billion-parameter open-weight model with its refusal behaviour removed through abliteration, so it answers instead of lecturing. It thinks before it answers, remembers the last 80 messages and writes up to ~2,500 words per reply.

Is Helios 1 uncensored?

Yes. Abliteration finds the single direction inside the model that makes it refuse and removes it from the weights, so it answers the question you actually asked. It runs on GPUs we operate ourselves, and your prompts aren't sent to another AI company. A few hard limits still apply, listed in our guidelines.

How do I use Helios 1?

Helios 1 is included in the Max and Max 20x plans, where it's the default model. Developers can also call it through the Bleepless API with the model id helios-1.

Our other models.