Day 7 · 2026-10-30
15 launches checked60 claims quoted word for word87 user reports linked10 verified fixes linked6 corrections publishedLive 7 days5 of 5 founding sponsor spots left
Day 7Launch day vs. day 7
AI launches, checked against real usersLast re-stamped 30 Oct 2026

AI models / Compare

EmbeddingGemma 2 vs LFM3: which delivered after 7 days?

EmbeddingGemma 2 is stamped Delivered; LFM3 is still Too early to call, so this comparison isn't settled yet.

EmbeddingGemma 2LFM3
VerdictDeliveredToo early
Claims held40
Claims mixed10
Claims failed00
Claims unclear03
User reports142
Launched6 Oct 202627 Oct 2026
Last stamped25 Oct 202629 Oct 2026

No. 009 · Google

EmbeddingGemma 2

DeliveredDay 7 verdict

The open weights, the shared text-image-audio space and the code gains check out in independent tests; text-only users get no boost, as Google said.

Launched
6 Oct 2026
Stamped
25 Oct 2026
Claims
✓4 held · ~1 mixed · ✗0 failed · ?0 unclear
Reports
14 (6 praise · 5 problem · 3 mixed)

What they claimed

“Today, we’re launching EmbeddingGemma 2, expanding beyond text to unify code, images, video, and audio in a shared embedding space.”
✓ Holds
“Built on the Gemma 4 architecture and released under a commercially permissive Apache 2.0 license, EmbeddingGemma 2 has 740 million parameters, making it optimal for on-device inference.”
✓ Holds
“Using Matryoshka Representation Learning (MRL), developers can dynamically truncate output vectors from 768 dimensions down to 512, 256, or 128 dimensions.”
✓ Holds
“EmbeddingGemma 2 matches the strong multilingual text performance of EmbeddingGemma while delivering a significant 9.92-point improvement on code performance (in MTEB Code, from 68.76 to 78.68), making it well-suited for local codebase indexing, semantic code search, and coding agent retrieval.”
✓ Holds
“We worked closely with the following partners to ensure EmbeddingGemma 2 works immediately where you build:”
~ Mixed

Full receipt

No. 013 · Liquid AI

LFM3

Day 3 of 7
Too earlyDay 7 verdict

Two days in, only a few first-hand comments exist, and one reports a load failure and a wrong answer.

Launched
27 Oct 2026
Stamped
29 Oct 2026
Claims
✓0 held · ~0 mixed · ✗0 failed · ?3 unclear
Reports
2 (0 praise · 1 problem · 1 mixed)

What they claimed

“a new tri-hybrid architecture and training recipe allow LFM3-1.7B to outperform our previous LFM2.5-2.6B while using 35% fewer parameters.”
? Unclear
“the text model fits in about 1 GB at 4-bit precision.”
? Unclear
“256K context window for the text model.”
? Unclear

Full receipt

Both stamps come from the public user reports linked on each receipt, under the same method. Verdicts are opinion and are never paid for.

Keep comparingall AI models