INTRODUCING SPHOTAA state-of-the-art 240M parameter language model
INTRODUCING

Sphotasphoṭaḥ

A meaning-first
language model.

A language model built around a representation field, connecting context with expression. The first release from a different approach to intelligence.

Explore performance
TOKENSMEANING FIELDEXPRESSION
01 / THE ARCHITECTURE ADVANTAGE

Intelligence, with
less overhead.

01 / COMPUTE
3.39×

Compute efficiency.

Compute efficiency relative to Qwen3-0.6B under the stated workload.

VS QWEN3-0.6B
02 / TRAINING
~2T

More from each token.

Trained on fewer tokens. Outperforming competitors trained on 28T.

TRAINING TOKENS
03 / MEMORY
86.7%

Lower memory use.

7.54× memory efficiency relative to Qwen3-0.6B.

VS QWEN3-0.6B
04 / CAPABILITY
<1B

State of the art.

240M parameters. Ahead of LFM 700M in the evaluated comparison.

PARAMETER CLASS

EFFICIENCY WORKLOAD32,768 input tokens + 256 output tokens · BF16 · batch 1 · KV-cached decoding.

02 / EVALUATIONS

Performance,
in perspective.

15 TASKS · 5 MODELS
REPORTED TASK SCORE / HIGHER IS BETTER

IFEval p-s

Task leader Sphota 240M

79.10SPHOTA · TASK SCORE

Instruction following · 1 of 15 tasks.
Each bar shows this task’s reported score.

Each stage compares the reported scores for one task. The final Overall stage uses the supplied summary scores.

03 / DEMONSTRATION

See Sphota
Think.

RECORDED CPU OUTPUTSTHREE SELECTED EXAMPLES
ONE QUESTION / FOUR MODELS

Following the full chain.

Maya is the parent of Leo. Ben is the parent of Alice. George is the parent of Jacob. Alice is the parent of David.

QUESTION

Who is the grandparent of David? Answer with only the name.

EXPECTED ANSWERBen

Sphota240M

Mannas

Ben

Correct · matches expected answer

LFM2.5350M

Liquid AI

David

Does not match

LFM2700M

Liquid AI

George

Does not match

Qwen3600M

Alibaba

David's grandparent is Alice.

Does not match

Actual CPU benchmark outputs; each model received the same question.

Benchmark settings

Greedy decoding · BF16 · 8 CPU cores · batch 1 · max 48 new tokens · no web search · no added system prompt.

Native Hugging Face chat templates for the baselines; Qwen thinking off. Sphota uses a Question/Answer wrapper.

EARLY ACCESS

Put Sphota
to work.

Explore what a different approach to language can make possible.

Mannas / CONTACT

Get in touch.

A research question, a collaboration, or something you’d like to explore.

Opens a draft in your email app, addressed to hello@mannas.ai. Nothing is sent automatically.