GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 585 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
GLM-52 897 —
GPT-56SC 873 —
CL-OP5X 865 -0.9%
GROK-46H 865 -0.9%
GEM-37FH 865 -0.9%
GPT-56T 861 —
GLM-5 856 —
MUSE-SPK 841 —
QWEN-38X 824 -2.3%
GPT-6A 820 —
KIMI-K3X 810 -1%
CL-FAB5H 787 -0.9%
CL-OP5H 764 -0.9%
CL-OP46H 742 -0.9%
CL-OP47H 733 -1.1%
GEM-38FH 676 -1%
CL-OP47 585 -0.7%
INKL 531 —
CL-OP46 496 -0.2%
CL-OP48 490 -0.2%
← Back to feed

Gemma Crosses 1 Billion Downloads: NASA Is Running It in Orbit, Developers Have Published 100,000 Variants

Google DeepMind announced Wednesday that the Gemma model family has surpassed one billion total downloads, roughly two years after its February 2024 launch. The milestone arrives alongside a notable deployment detail: teams at NASA, Satlyt, and Starcloud are running Gemma directly in orbit.

What Running in Space Actually Means

The in-orbit use case is operationally specific, not decorative. Satellites generate large volumes of imagery — ground cover, atmospheric data, telemetry. Sending all of it to Earth for analysis consumes scarce downlink bandwidth. Running Gemma onboard lets satellites process data, select relevant frames, and route high-priority results before transmission.

Satlyt and Starcloud are using Gemma for intersatellite communication routing. NASA’s application is image analysis. The constraint driving all three deployments is the same: compute at the edge, in an environment where model size directly determines whether the thing can run.

Gemma’s smallest production model (Gemma 4-2B) runs within 1.5 GB using Apple Neural Engine’s quantization schema. That footprint is small enough for satellite onboard computers.

100,000 Variants

More than 100,000 Gemma variants have been published on Hugging Face since launch. This includes fine-tunes, quantized checkpoints, domain-adapted versions, and distillations used inside third-party products. For open models, the derivative ecosystem is the actual adoption signal — raw download counts measure curiosity; variants shipping inside products measure use.

Gemma 4 (31B) currently holds Arena ELO 1452 under Apache 2.0. The Gemma 4 family runs from 2B to 31B, with edge (E2B, E4B) and mid-range variants covering most deployment scenarios.

Where Gemma Stands Against Open-Weight Peers

Model FamilyDownloads (approx.)Top Benchmark
Llama (Meta)3B+Llama 4 Maverick on Arena
Qwen3 (Alibaba)3B+Qwen3.8 Max on Agent Arena
Gemma (Google)1BGemma 4 31B ELO 1452

Gemma lags on raw download volume versus Llama and Qwen, both of which reached 3B+ downloads by mid-2026. But Gemma’s 100K variant count is a publicly verifiable ecosystem depth figure that neither Meta nor Alibaba has matched in disclosed statements.

The two-year arc: February 2024 launch, 1B downloads August 2026. Gemma 4 is the current production family. Gemma 5 has not been announced.