Gemma 4 E4B is the pick; LFM2.5 8B-A1B gained the most

Pick of the week

8 billion on disk, under a gigabyte of context at 32K

Gemma 4 E4B · Google · Q4_K_M 4.6 GB

Gemma 4 E4B by Google takes the eighth pick: Heat Score 59 to 76 between the readings of 24 and 30 October, the largest gain among models that run comfortably on a 24 GB card and are not on cooldown. Its Q4_K_M file measures 4.6 GB. At 8K context it needs 5.8 GB, it runs on the 8 GB cards and runs easily from 10 GB up; at 32K it needs 7.2 GB, and only 0.9 GB of that is context cache, a fifth of what last week's pick needs at the same length.

The pick on your card

How Gemma 4 E4B's memory grows with context

The model accepts up to 131,072 tokens of context. Memory = measured weights + computed context cache + runtime, on an RTX 4090's 21.1 GB usable.

What to run this weekend

The largest and the hottest model that fit each card class

"Runs" means our verdict for the measured Q4_K_M file at 8K context is "Runs easily" or "Runs". Largest = most parameters; hottest = highest popularity score. Check your own card

Departures board

Who gained and who lost this week

The score behind these moves is the one the board shows: a published composite of Hub trending, download growth, 30-day downloads and likes. On the pages it reads as popularity. The full board

Price board

What an hour cost last week

Week median = the median of the week's daily medians, glitch days excluded. "Few offers" marks a median over fewer than five offers. Cloud vs. market = cheapest big-cloud list price ÷ the same day's Vast.ai median. What these numbers are

Chart of the week

RTX 4090: fourteen days of hourly prices

The RTX 4090 made the week's largest move on the price board, −36 % against the week before.

Daily median of Vast.ai's verified offers, UTC days, glitch days excluded by the same rule as the price board. Gaps are days without a reading.

Price moves

The week's largest day-over-day moves

• 31 Oct · A100 at Clore.ai: +59 % ($0.38 → $0.61)
• 28 Oct · A100 at Clore.ai: −29 % ($0.58 → $0.42)
• 28 Oct · H200 at Vast.ai: +25 % ($5.05 → $6.30)

A number explained

Five cards, five sources, no basket

A reference price that settles contracts weighs five cards: H100 at 40 percent, B200 at 25, H200 at 15, A100 at 10 and RTX 4090 at 10. It publishes one number. This board publishes the five cards, source by source, and no number above them. On Sunday 1 November at 20:00 UTC one hour was listed at: H100 $2.59 posted at RunPod, $2.27 typical on the Vast marketplace across six offers, $3.29 at Lambda, $2.50 at Hyperstack, $6.88 at AWS. B200 $5.98 at RunPod, $8.75 on the Vast marketplace across six offers, $6.69 at Lambda, $6.00 at Hyperstack, $14.24 at AWS.

H200 $3.59 at RunPod, $5.13 on the Vast marketplace across eight offers, $3.99 at Hyperstack, $7.91 at AWS; Lambda does not list it. A100 $1.19 at RunPod, $0.84 on the Vast marketplace across 41 offers, $1.99 at Lambda, $1.35 at Hyperstack, $2.74 at AWS. RTX 4090 $0.34 at RunPod and $0.47 on the Vast marketplace across 64 offers; no big cloud lists it.

Read across a row and the spread is the story: for an H100 hour the cheapest and the dearest listing differ by a factor of three, for a B200 by 2.4, for an A100 by 3.3. A marketplace middle over six offers is a reading, not a market; a cloud's list price buys a whole instance with its own terms. Which of these a basket takes, and how a paid price differs from a listed one, is the difference between an index and a board. We do not make an index.

Pick of the week 8 is Gemma 4 E4B: it grew in popularity faster than any other model that fits a 24 GB card and was not picked in the last four weeks. Its file measures 4.6 GB; 7 of its 42 layers cache the full context, so it needs 7.2 GB at 32K and runs easily on a 12 GB card. The week's board read: the RTX 4090 hour fell by a third against the week before, $0.50 against $0.76 as week middles on the Vast marketplace, and the H100 hour rose, $3.08 against $2.40, on six to eight offers. Every daily row behind these numbers is a CSV download, up to a year back, CC BY 4.0.

What the pick costs

Gemma 4 E4B, two ways, per day

For 500,000 tokens a day over 4 active hours, the calculator's "regular" preset:

Newly measured

Files measured this week, and where they fit

• Gemma 4 E4B (Q4_K_M, 4.6 GB): fits a 12 GB, 16 GB, 24 GB card comfortably at 8K.
• Ling 3.0 Tiny (Q4_K_M, 4.5 GB): fits a 12 GB, 16 GB, 24 GB card comfortably at 8K.
• Phi-4 Mini (Q4_K_M, 2.3 GB): fits a 12 GB, 16 GB, 24 GB card comfortably at 8K.
• Qwen3 4B (Q4_K_M, 2.3 GB): fits a 12 GB, 16 GB, 24 GB card comfortably at 8K.
• Granite 4.2 3B (Q4_K_M, 2.1 GB): fits a 12 GB, 16 GB, 24 GB card comfortably at 8K.
• Qwen3-Coder 30B-A3B (Q4_K_M, 17.3 GB): fits no consumer card comfortably at 8K.

Memory from the measured GGUF file plus the computed context cache; "runs" means our verdict is "Runs easily" or "Runs". Check your own card

Rent or buy, one number

3,263 rented hours

At last week's Vast.ai median of $0.49 an hour, the RTX 4090's launch list price of $1,599 (NVIDIA launch MSRP, Oct 2022) equals 3,263 rented hours, ≈ 8 years at 8 h/week. Electricity and resale value are left out on both sides.

Class of the week

L40S

Last week's Vast.ai median: $0.80 an hour. Since we started reading it 28 days ago: 0 %. Cheapest source at the week's end: RunPod at $0.79 (list price). The cheapest big cloud, AWS, listed 2.3× the marketplace median. 30 of 48 tracked open models run comfortably on one at 8K context; the largest comfortable one is Qwen AgentWorld 35B-A3B.

Maker board

Which labs hold the top-10 slots

Counted from the popularity ranking when this issue was built, Monday 05:45 UTC; the 7-day change is the mean per-model change. The full board

New on the radar

Added this week

• Ornith 1.5 122B (Ornith AI), added 28 Oct
• LFM2.5 24B-A2B (Liquid AI), added 28 Oct
• Radar Cards claimed: 5 in total. The next number of the RTX 4090 card is #2.

From the tower

What changed on the site

• On cooldown and therefore not eligible as the next pick: Gemma 4 E4B (picked 2026-w44), Qwen3 8B (picked 2026-w43), LFM2.5 8B-A1B (picked 2026-w42), Ornith 1.5 9B (picked 2026-w41). The rule

From the methodology desk

What this issue does not say

Prices above were read on Sunday 1 November 2026 at 20:00 UTC; marketplace figures are the middle of the offers listed then, with the offer count, and cloud figures are list prices for one GPU. Six or eight offers behind an H100 or B200 middle make it a reading, not a market. The index weights are stated as a fact about a third party's method; we neither publish that index nor link to products built on it. The two models added to the catalog this week, LFM2.5 24B-A2B and Ornith 1.5 122B, have no popularity reading before 4 November. No forecast.