GribStream

GribStream Blog

ECMWF confirmed the May 12, 2026 IFS Cycle 50r1 and AIFS v2 upgrade

|
ifsoper ifsenfo ifswave ifswaef aifsoper aifsenfo

ECMWF implemented IFS Cycle 50r1, AIFS Single v2, and AIFS ENS v2 from the 06 UTC run on May 12, 2026, with same-day model, stream, wave, and level changes to check.

ECMWF confirmed and then implemented one of its larger coordinated model upgrades of 2026: IFS Cycle 50r1, AIFS Single v2, and AIFS ENS v2 all moved at the 2026-05-12 06 UTC run.

The important point for users is that ECMWF's physics-based forecast system and AI forecast systems changed together. If you compare IFS versus AIFS, or use either one as a reference for verification, the same timestamp is a model-version boundary on both sides of the comparison.

GribStream follow-up posts cover what changed in the public data we serve:

What ECMWF implemented on May 12

ECMWF's forecast-change pages now show the coordinated implementation:

  • IFS Cycle 50r1: implemented from the 2026-05-12 06 UTC run
  • AIFS Single v2: implemented from the 2026-05-12 06 UTC run
  • AIFS ENS v2: implemented from the 2026-05-12 06 UTC run

The same-day implementation matters because AIFS v2 was trained and tuned around the new IFS cycle. ECMWF explicitly says AIFS Single v2 needed to move with 50r1 because AIFS Single v1.1 degraded when initialized from 50r1 esuite data.

What the official pages actually say changes

The 50r1 implementation page says the upgrade applies to the IFS medium-range forecast, including both the atmospheric and wave components.

For users, the most concrete details are:

  • ECMWF says there is no change in horizontal or vertical resolution, and no change in steps, for either the atmospheric or wave model in 50r1
  • the ENS control archive/dissemination handling changes, with control forecasts moving under the operational stream
  • the wave control handling changes too, with wave control forecasts moving under the wave stream

No grid change is planned for 50r1. The operational impact is the same-day model update plus stream/archive changes for direct ECMWF users.

ECMWF's newsletter overview is still the best short summary of the science side of 50r1:

  • the ocean and sea-ice core moves to NEMO4-SI3
  • coupled data assimilation becomes more central
  • several physics and assimilation changes target humidity, wind, and stratospheric behavior
  • ECMWF also highlighted major compute-efficiency gains in the 50r1 design

AIFS-specific changes worth checking

The AIFS implementation pages add several concrete details:

  • both AIFS Single v2 and AIFS ENS v2 were implemented from the 2026-05-12 06 UTC run
  • both add a new 10 hPa pressure level to the stratospheric component
  • AIFS Single v2 adds a new wave component and wave stream
  • AIFS ENS v2 also introduces a new wave stream
  • AIFS Single v2 changes vertical velocity w from a prognostic field to a diagnostic field
  • ECMWF points to new snow and wave variables in the AIFS v2 product set

If you ingest AIFS fields directly, those are the details to check first. Some of those official ECMWF additions are now visible in GribStream's public AIFS catalogs, while others depend on which ECMWF public open-data streams are actually present and stable enough for GribStream to serve.

What to check

If you run verification, threshold-based alerts, or model-to-model comparisons, split your analysis around 2026-05-12 06 UTC.

That is especially important if you:

  • compare IFS against AIFS
  • maintain marine workflows that depend on the wave side of IFS
  • use long history windows and assume one stable model behavior through spring 2026
  • pull ECMWF data directly from MARS or dissemination streams and care about stream names or control-product handling
  • compare AIFS v1/v1.1 output against AIFS v2 output

It is also worth keeping this separate from the later 50r2 GRIB2 migration. That is a different transition with different failure modes. May 12 is about the model-cycle upgrade itself, not the later all-GRIB2 cutover we covered in our 50r2 post.

Related reading

Sources

Related models

ifsoper ifsenfo ifswave ifswaef aifsoper aifsenfo
Try GribStream