The Spread

Where the models disagree today

26 archived cities · daily high temperature · day 3 (2026-09-08) · guidance as of 2026-09-05T06:50Z

Three days out, to Monday the 8th, the forecasts computers run split hardest in Minneapolis: across our set of 26 archived cities, the seven models there range from a cool 75 to a warm 97 for the daytime high, and they don't cluster around one answer but scatter into four separate groups. San Francisco is nearly as unsettled — the models sit anywhere from 65 to 84 and fall into five distinct camps. When the models spread that widely and clump into different groups, it means no single forecast for that day should be treated as settled; the true answer could plausibly land at either end.

By contrast, several cities are in near-agreement for Monday. Phoenix (around 106), Philadelphia (around 84), Charlotte and Denver all show the models within four to five degrees of one another and, mostly, telling one story. A forecast you read for those places today is more likely to hold.

Two cities shifted noticeably in the last day. The models' middle-of-the-road high for Boston on Monday climbed five degrees, now landing near 81, and San Francisco's rose about five to near 75. That kind of overnight jump is itself a signal the picture is still moving. A related note: the model groupings for San Antonio, Oklahoma City, Dallas, Houston, New Orleans and Atlanta all realigned in the same direction over the past day, tightening toward agreement in that band of southern cities.

Across every city and lead in the archive, the American GFS model runs the warmest of the seven, averaging a bit over a degree above the group and posting far more hot-side outliers than any other. The European AIFS leans coolest. Worth remembering that these are days-ahead computer runs, not measurements — the further out the day, the wider these gaps tend to open, as the seven-day columns show plainly.

Issued 2026-09-05T00:58Z from 26 cities' archived model runs. Written from the numbers on this page — no forecast of its own.

Where they're fighting

Each marker is one archived city, placed geographically. The label gives the coolest and warmest model forecast for day 3 — so a wide pair is a city the models disagree about — and marker size is the gap between them. The coastline, lakes and state lines are there to place them; no value is drawn between the cities, because we know these 26 points and not the country between them.

Seattle 69–74° Minneapolis 75–97° Boston 72–82° Buffalo 74–81° Detroit 72–84° New York 80–86° Chicago 82–93° Salt Lake City 75–83° Philadelphia 81–85° San Francisco 65–84° Washington 82–87° Denver 84–88° St. Louis 90–100° Los Angeles 86–93° Charlotte 83–88° Oklahoma City 99–106° San Diego 84–98° Phoenix 105–109° Atlanta 81–89° Dallas 97–102° New Orleans 85–93° Houston 90–98° San Antonio 95–102° Orlando 86–101° Tampa 86–95° Miami 85–92°

A filled marker means the models fractured into three or more distinct camps there — not just a wide range, but several competing answers at once. Most cities carry two camps at this lead; these are the ones telling more stories than that.

These markers are day 3. For today's spread in each of these cities, see every location we archive.

The maps

The national field maps have their own pages — one map each, so a page carries one coastline instead of four.

Which models run hot

Each model's average distance from the cross-model median, across every archived city and lead. Positive runs warm. Outlier columns count model-days at least 3° off the median. Consensus products are excluded — they are averages of the rows below, not competitors.

Modelvs median WarmCool Model-days
GFS +1.63° 29 11 78
UKMO +0.42° 7 3 52
ECMWF +0.28° 15 12 78
AIGFS -0.02° 13 8 78
ICON -0.11° 2 3 52
GEM -0.51° 6 10 78
AIFS -1.16° 6 19 78

What moved overnight

Cities whose cross-model median for 2026-09-08 shifted at least 3° in the last 24 hours.

  • Boston +5° → 81°
  • San Francisco +5° → 75°

How this page is built

Every number here is computed from The Spread's own archive of what each model actually published, run by run. The written summary at the top narrates those numbers and is not allowed to add any — no fronts, no systems, no forecast. It describes what the models are doing, which is what this archive measures.

Sample is 26 archived US cities, so "everywhere" means those cities. The map draws markers only: interpolating between them would show a pattern nobody measured.