How much does TOAI-elegant-lang reduce token use?

On internal inter-agent traffic it reports -89.4% to -89.5%, a 21.7x compression, based on over 2,090 production records. For Japanese A2A packets the reported cut is -89.8% (9.8x).

Which AI models were tested with the protocol?

Ten models from nine labs were connected via API, including Google DeepMind's gemini-3.5-flash-lite, OpenAI's gpt-6.1-sol and gpt-6-astra, Anthropic's claude-sonnet-5.5 and claude-opus-5.5, and xAI's grok-4.7.

Can I try it for free?

Yes. The Open Core Edition, including the 256-word RISC base dictionary, rosetta_layer.json, Python SDK, CLI and MCP server, is released under the MIT License at no cost.

Large Language ModelsOpen-Source AIQiita 機械学習5h agoOct 28, 2026, 13:00 JST

Hinotori releases TOAI-elegant-lang, claiming 21.7x AI-to-AI token cut

Send on LINE

3 Key Points

What happened
The Hinotori guild released TOAI-elegant-lang under an MIT-licensed Open Core edition, reporting a 21.7x compression ratio (-89.4% to -89.5%) on internal inter-agent traffic from over 2,090 production records.
Why it matters
If the measured savings hold across companies, AI agents talking to each other could consume far fewer tokens, cutting the compute and power behind multi-agent workflows.

WHO IT HITSEnterprise teams running multi-agent systems or Japanese-language agent traffic stand to cut token bills and compute load, though the largest savings are from the author's own harmonized logs.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

The protocol is meant to solve a specific waste: AI agents still exchange messages in human language, generating verbose text that the receiving model must re-parse token by token. TOAI-elegant-lang instead compresses control flow into 1–2-token ASCII identifiers, while keeping practical parameters such as numbers and currency pairs fully intact. The project also targets Japanese, which it says consumes 1.3–1.8x more tokens than English due to subword splitting. To address fears of opaque AI collusion, the design centers on a Rosetta-stone-style three-script architecture: a readable concept layer, a highly compressed wire layer, and a deterministic dictionary that can restore every symbol to Japanese or English. The claimed result is a black-box rate of 0.0%, with the dictionary and SDK published openly. The reported benchmark ratios vary by model, ranging from 3.62x for Alibaba's qwen3.8-flash to 10.96x for Zhipu AI's glm-5.3-flash. The project is offered as open core: the base dictionary and tools are MIT-licensed for free, while production use of the higher-level macro dictionary and compression middleware requires sponsorship.

FAQ

How much does TOAI-elegant-lang reduce token use?: On internal inter-agent traffic it reports -89.4% to -89.5%, a 21.7x compression, based on over 2,090 production records. For Japanese A2A packets the reported cut is -89.8% (9.8x).
Which AI models were tested with the protocol?: Ten models from nine labs were connected via API, including Google DeepMind's gemini-3.5-flash-lite, OpenAI's gpt-6.1-sol and gpt-6-astra, Anthropic's claude-sonnet-5.5 and claude-opus-5.5, and xAI's grok-4.7.
Can I try it for free?: Yes. The Open Core Edition, including the 256-word RISC base dictionary, rosetta_layer.json, Python SDK, CLI and MCP server, is released under the MIT License at no cost.

Qiita 機械学習Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.