中文
~/home / Show HN
Show HN

Maple-Preview – Ternary 20B MoE running at 120 tok/s on a iPhone

Maple-Preview – Ternary 20B MoE running at 120 tok/s on a iPhone is a general AI agent for Run large language models locally on an iPhone for fast, private AI reasoning without relying on cloud services. Pricing: Free. As of 2026-09-20, KanonAgent records 173 upvotes. First indexed by KanonAgent on 2026-08-04.
Runs a 120-billion-parameter ternary MoE model at 120 tokens per second directly on an iPhone, enabling on-device AI inference with low latency and strong privacy.
173 upvotes
Tracked by Kanon since Aug 4, 2026
Breaking out
status
13/100
momentum · conf 0.47
28d
tracked since 2026-08-22
2026-08-05
last meaningful change
1
evidence records

No signal event on file yet — status is unknown, not "quiet".

📈 Timelinewhat changed, and when
🔎 Known / Unknown 15 of 23 fields unknown
Machine-callable unknown
Open source unknown
Self-hostable1 inferred
Bring your own key unknown
Autonomy level unknown
Pricing model unknown
Integrations unknown

“Unknown” means we have not verified it — it is not a “no”. Hard filters never treat unknown as false.

🚀 Breakout logappend-only · timestamped
2026-08-05Hacker Newsflagged at 120 points → now ▲173
Full breakout log →
🤖 Agent teardown · general
Job to be doneRun large language models locally on an iPhone for fast, private AI reasoning without relying on cloud services.
Who it is forMobile developers and edge AI users
Prerequisitesself-hostable
PricingFree
Traction · why it is risingGained 20 votes on Product Hunt
Why it matters

Breaks performance barriers for on-device AI, making large models practical on consumer hardware and paving the way for real-world edge AI applications.

Evidence quotesverbatim, from the product’s own materials

“Running a 120B ternary MoE model at 120 tok/s on an iPhone”— description
Signal source: Show HN
Visit official site →
📛 Official badgefor your site / README
Maple-Preview – Ternary 20B MoE running at 120 tok/s on a iPhone badge
Building Maple-Preview? Pick a style above — the embed code updates live. Deep color control via URL params: bg= / fg= / accent= (hex). It links back to this page.
Share on X
On mobile tap Share for WeChat / RED (Xiaohongshu) / X; on desktop use Copy text and paste into the app.

FAQ

What is Maple-Preview?

Runs a 120-billion-parameter ternary MoE model at 120 tokens per second directly on an iPhone, enabling on-device AI inference with low latency and strong privacy.

What does Maple-Preview do?

Run large language models locally on an iPhone for fast, private AI reasoning without relying on cloud services.

Why does Maple-Preview matter?

Breaks performance barriers for on-device AI, making large models practical on consumer hardware and paving the way for real-world edge AI applications.

How much does Maple-Preview cost?

Free

Is Maple-Preview free?

Yes — Maple-Preview has a free tier. Pricing as stated on its own page: Free

Can I self-host Maple-Preview?

Yes — Maple-Preview can be self-hosted.

How popular is Maple-Preview?

As tracked by KanonAgent: 173 upvotes (first indexed 2026-08-04).

When did Maple-Preview take off?

KanonAgent flagged Maple-Preview as breaking out on 2026-08-05 on Hacker News (at 120 points that moment). The breakout log is append-only: https://kanonagent.com/rising

What are the best Maple-Preview alternatives?

Similar AI agents tracked by KanonAgent: Stealth Company, hermes-agent, Anonymous Startup, AutoGPT, Anonymous Startup, Confidential Startup.

Maple-Preview – Ternary 20B MoE running at 120 tok/s on a iPhone alternatives — similar AI agents

Stealth Companyhermes-agentAnonymous StartupAutoGPTAnonymous StartupConfidential Startup

Where this fits — browse the same shelf

Autonomous AI AgentsAI agents for Local deployment and runningCan I self-host an AI agent for Local deployment and running?Self-hosted & open-source AI agents