Underclassic rss archive buyer's guide
Issue 107 · 2026-11-17

Permanent copy of issue 107. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Edge0 has released Edge0-145B, an open-weight model with 145 billion parameters that the company says peaks at about 29 GB of memory and generates more than 80 tokens a second on a laptop such as a MacBook Pro. No independent tester has yet confirmed those figures, but if they hold, work that now requires a rented cluster could run on a machine a household already owns, on the same day Nvidia reports its first results since its chief executive resigned and Washington moves to police who may rent its chips.
Date: 2026-11-17
Updated: 2026-11-17 06:30
Status: LIVE
Broadcast
Edge0's claim is simple to state and still unproven: a model with 145 billion parameters, running on one laptop, at a speed faster than a person reads. If outside testers confirm it, a good part of the coding and research work now sold by the token could move onto machines people already own. The clusters are not standing still. PJM wants to connect them in a year instead of two, while Rochester has joined the towns that pause them first and write the rules later, and Nvidia faces its quarterly results with no permanent chief executive and new scrutiny from the Treasury and Congress.
Watch
Stories
Edge0 has published the weights for Edge0-145B, a mixture-of-experts model derived from Alibaba's Qwen3.8-Flash-Next, and says it peaks at 29 GB of memory and decodes at more than 80 tokens a second on a compatible laptop such as a MacBook Pro. Those figures and its coding benchmark scores are the company's own; reviewers found no independent tests by Nov. 14, and Edge0's GitHub instructions, which at launch documented only Nvidia GPUs on Linux or Windows for the 145B model, now also list Apple Silicon Macs.
-- A model of this size that runs on one personal machine would let a household or small firm do coding and research work offline with no per-token bill; our Open-Source AI Guide lists it in the gaming PC tier at /open-source-ai.html until outside tests confirm the speed.
See also: Underclassic model board: Edge0-145B next to cloud list prices
2026-11-12 / Edge0 / efficiency / local / #edge0-releases-a-145-billion-parameter-open-mode
PJM Interconnection said it filed its Fast Energization proposal with the Federal Energy Regulatory Commission on Monday, studying new generation, transmission and large loads such as data centers together and cutting the projected study time to one year from two. The plan moves readiness and financial requirements earlier to weed out speculative projects, and PJM asked FERC for a 60-day comment period and an order within 120 days.
-- PJM runs the grid for 13 states and Washington, D.C.; a faster queue helps large clusters connect, while its promise to show states the cost of upgrades driven by data centers decides whether households or the data centers pay for them.
2026-11-16 / PJM Inside Lines / energy / central / #pjm-asks-federal-regulators-to-halve-the-wait-fo
The Rochester City Council voted 5-1 on Monday for an interim ordinance that halts data centers within city limits for a year and orders a study of best practices, KTTC reports; eight of the ten residents who spoke at the hearing supported it, and the city says no applications are pending. The Underclassic Data Center Fights Tracker counts 46 of 47 local ballot measures on data centers this year as limiting development, with Hollis, Maine, voting on a 180-day moratorium on Dec. 8.
-- Like most recent pauses, Rochester's arrives before any developer has filed, so builders of large clusters face rules written in advance rather than negotiated project by project.
See also: Data Center Ballot Results 2026: every local vote, pause and hearing
2026-11-16 / KTTC; Underclassic Data Center Fights Tracker / energy / local / #rochester-minnesota-approves-a-one-year-pause-on
Treasury Secretary Scott Bessent called on Sunday for a "dramatic shakeup" at Nvidia after Jensen Huang's resignation and said the board should look outside the company for his successor, CNBC reports. Bessent questioned whether Nvidia pressed sales that may have breached export controls to lift earnings; Nvidia reports its first quarterly results under interim chief executive Colette Kress on Tuesday.
-- Nvidia supplies most of the chips in large AI clusters, so a government-pressed change of leadership could slow its overseas sales just as the company must show investors that demand is intact.
2026-11-15 / CNBC / chip / mixed / #bessent-urges-nvidia-s-board-to-hire-an-outside-
Rep. Carlos Gimenez said on Nov. 15 that the United States must close the cloud and third-country channels that let Chinese firms use advanced AI processors, East Asia Brief reports, building on the Remote Access Security Act the House passed 369-22 in January. The chairs of three House committees have summoned Nvidia general counsel Tim Teter and Commerce Under Secretary Jeffrey Kessler to testify and want the measure in the annual defense bill.
-- Treating remote access as an export would require cloud operators to verify who rents their clusters, a compliance cost that falls on central providers and not on someone running a model on their own machine.
2026-11-15 / East Asia Brief / chip / mixed / #house-committee-chairs-push-to-extend-chip-expor
Axis
local [########################------] central
Score: 79
clusters still winning
-1
Tuesday's edition leads with Edge0-145B, an open model its maker says runs on a laptop, alongside PJM's plan to connect data centers faster, a one-year moratorium in Rochester, Minnesota, and new pressure on Nvidia from the Treasury and Congress. A credible large model for personal machines is the strongest local signal in weeks, though its numbers are still the company's own, so the score eases one point to 79.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
96 | central | Treasury Secretary Scott Bessent has urged Nvidia's board to hire an outside chief executive, and House committee chairs want export curbs extended to rented cloud access. Nvidia reports quarterly results on Tuesday; the gauge holds at 96. | ||
| ENERGY |
|
95 | central | PJM asked federal regulators on Monday to halve the time it takes to study connections for data centers and power plants, while Rochester, Minnesota, paused new data centers for a year. One measure speeds clusters onto the grid and the other slows them, so the gauge holds at 95. | ||
| EFFICIENCY |
|
88 | local | Edge0 released Edge0-145B, a 145-billion-parameter open model it says runs in about 29 GB of memory on a personal computer. The claims are self-reported, so the gauge rises only one point, to 88. |
Line charts (0=local 100=central 2026-11-02 to 2026-11-17)
Market overlay (2026-10-19 to 2026-11-17)
Latest: BTC $64,155 / hashrate ~972.06 EH/s / token markers avg $5.2583/M / axis 79. These lines are different units. We compare timing, not dollars-to-hashrate.
Not shown: GPU cloud $/hour, hub electricity, miner hashprice, or lab cost per token -- none of those are a clean public feed.
Power and list prices
BTC $64,155 / hashrate ~972 EH/s. List-price gap (not a subsidy %): 123.1x list gap (blended) Mercury 3 -> Claude Fable 5.1 ($0.16/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)
Marker prices (USD per 1M tokens; verified 2026-11-15; blended = (3*input + 1*output) / 4; -- = open/self-host or no single public list)
| # | Model | Org | CC | Weights | Tier | In | Out | Blend |
| 1 | Claude Sonnet 5 | Anthropic | US | closed undisclosed |
workhorse | $2.00 | $10.00 | $4.00 |
| 2 | GPT-5.5 | OpenAI | US | closed undisclosed |
flagship | $5.00 | $30.00 | $11.25 |
| 3 | DeepSeek V4.1 Flash | DeepSeek | CN | open ~284B total / ~13B active (MoE) |
commodity | $0.30 | $1.20 | $0.53 |
If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
Get the daily issue
One email a day. Unsubscribe anytime.
home | rss | page 2 | page 3 | archive | models | buyer's guide | subscribe | data | board | about | mcp | llms.txt