A complete four-skill open-model stack fits under 4 GB, runs at reading pace on the cheapest hardware by community measurements, and costs about one US cent of electricity per study hour¶

Claim · All claims
Evidence · 1 study (3 entries) · design r1 · q1
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q1 i? The complete four-skill stack totals under 4 GB, inside the 8-16 GB of the cheapest new laptops. → Guo 2026 q1 i? Community measurements report a 3B model at roughly 4-9 tokens per second on an $80 single-board computer and 5-15 tokens per second on the entry laptop CPU class, at or above the roughly four words per second of silent reading. → Guo 2026 (2) q1 i? At the 2024 average US residential electricity price, a device drawing 15 W during study costs about a quarter of one US cent per study hour. → Guo 2026 (3)

Evidence¶

Guo 2026¶

Guo, Q., Tang, J., Huang, X., Lin, H.-Y., Zhong, Y., & Zhuang, X. (2026). LLMersion: A Local-First AI Agent Framework for Low-Cost Home Language Learning toward Educational Equity. https://arxiv.org/abs/2609.29672 doi:10.48550/arxiv.2609.29672

q1 · i? · design · r1

This is the article's feasibility synthesis (Section 5.5, Table 3), a descriptive sizing argument from published file and model-card figures, not a measured experiment. It states "The complete four-skill stack totals under 4 GB, inside the 8–16 GB of the cheapest new laptops." No effect size applies.

"The complete four-skill stack totals under 4 GB, inside the 8–16 GB of the cheapest new laptops."

Guo 2026 (2)¶

Guo, Q., Tang, J., Huang, X., Lin, H.-Y., Zhong, Y., & Zhuang, X. (2026). LLMersion: A Local-First AI Agent Framework for Low-Cost Home Language Learning toward Educational Equity. https://arxiv.org/abs/2609.29672. doi:10.48550/arxiv.2609.29672

q1 · i? · design · r1

The article reports attributed community throughput measurements, explicitly noting no peer-reviewed benchmark exists for LLM inference on the cheapest CPUs. It benchmarks against "silent reading averages 238 words per minute, about four words per second". Throughput figures are descriptive; impact is null.

"reproducible community tables report a 3B model at roughly 4–9 tokens per second on an $80 single- board computer and 5–15 tokens per second on the entry laptop CPU class"

Guo 2026 (3)¶

Guo, Q., Tang, J., Huang, X., Lin, H.-Y., Zhong, Y., & Zhuang, X. (2026). LLMersion: A Local-First AI Agent Framework for Low-Cost Home Language Learning toward Educational Equity. https://arxiv.org/abs/2609.29672. doi:10.48550/arxiv.2609.29672

q1 · i? · design · r1

The article's cost derivation at the 2024 average US residential electricity price of 16.5 cents per kWh: "a device drawing 15 W during study ... costs about a quarter of one US cent per study hour". The arithmetic is the authors' own from cited inputs; no effect size applies.

"a device drawing 15 W during study, the measured envelope of entry CPUs under inference (Shah, 2026), costs about a quarter of one US cent per study hour"

Discussion¶

Related Claims¶

• Fine-tuning and running CLST is feasible on a single workstation GPU, with measured training cost of about 6.8 hours and 19 GB peak memory on Algebra05 — related
• The ELEVATE server pipeline runs locally on two heterogeneous consumer machines, keeping LLM and TTS models fully resident in memory for low-latency inference — related
• The linearity bias heuristic enables robust alignment with low-dimensional embeddings on standard CPUs, lowering computational prerequisites — related
• Households pay substantial recurring sums for private supplementary language education, with Korea at 29.2 trillion won in 2024 and English the leading subject
• An ELEVATE prototype achieves near real-time end-to-end tutoring interaction on consumer-grade school hardware without cloud GPUs — related
• NF4 4-bit quantization reduces VRAM usage by 38.7% for Qwen-7B and 37.9% for DeepSeek-MoE while maintaining practical response times — related