Compare/Darwin-4B-David vs LFM2.5-VL

AI tool comparison

Darwin-4B-David vs LFM2.5-VL

Which one should you ship with? Here is the side-by-side panel verdict, pricing read, reviewer split, and community vote comparison.

D

AI Models

Darwin-4B-David

4.5B merged model beats Gemma-4-31B on GPQA — no training needed

Ship

75%

Panel ship

Community

Paid

Entry

Darwin-4B-David is a 4.5-billion-parameter model that achieves 85.0% on GPQA Diamond — outperforming Google's Gemma-4-31B (84.3%) at roughly 1/7th the parameter count. The kicker: it required no training whatsoever. It was built in 45 minutes on a single H100 using MRI-guided DARE-TIES model merging, a novel variant of the merge-and-trim technique. The MRI-guided approach uses activation analysis to identify which parameters in each source model are most critical, then applies DARE-TIES merging only to the high-value weight regions. This avoids the catastrophic interference that usually degrades merged models. The result is a small model that inherits the strengths of multiple larger predecessors without any of the compute cost of fine-tuning. For the AI community, this is a meaningful data point: model merging continues to close the gap with expensive training runs. Darwin-4B-David demonstrates that thoughtful merge strategies can extract benchmark-level performance from models that are a fraction of the size, making capable AI more accessible on consumer hardware.

L

AI Models

LFM2.5-VL

450M vision-language model that runs in under 250ms on edge hardware

Ship

75%

Panel ship

Community

Paid

Entry

Liquid AI just shipped LFM2.5-VL, a 450M-parameter vision-language model engineered from the ground up for edge deployment. Unlike most VLMs that require a beefy GPU in the cloud, LFM2.5-VL targets devices like the Snapdragon 8 Elite, NVIDIA Jetson Orin, and AMD Ryzen AI — hitting sub-250ms latency on-device without any cloud round-trip. This model builds significantly on its predecessor with four new capabilities: bounding box prediction (81.28 on RefCOCO-M), multilingual support across 8 languages, function calling, and improved instruction following. Those aren't just benchmark checkboxes — bounding box prediction means you can run visual grounding and object detection pipelines on a phone or robot without any server involvement. Liquid AI is the MIT-spun startup behind Liquid Foundation Models (LFMs), a non-Transformer architecture that delivers competitive performance at a fraction of the memory footprint. LFM2.5-VL is available free on HuggingFace and through Liquid's LEAP inference platform. For builders targeting on-device AI — robotics, mobile, embedded — this is one of the most practical releases of the month.

Decision
Darwin-4B-David
LFM2.5-VL
Panel verdict
Ship · 3 ship / 1 skip
Ship · 3 ship / 1 skip
Community
No community votes yet
No community votes yet
Pricing
Open Source
Open Weights
Best for
4.5B merged model beats Gemma-4-31B on GPQA — no training needed
450M vision-language model that runs in under 250ms on edge hardware
Category
AI Models
AI Models

Reviewer scorecard

Builder
80/100 · ship

45 minutes on a single H100 to beat a 31B parameter model? That's an extraordinary efficiency ratio. MRI-guided merging is a technique I'll be watching closely. If this holds up across more benchmarks, it fundamentally changes how teams should think about building capable small models.

80/100 · ship

Sub-250ms on-device vision with function calling is the unlock for a huge class of apps that couldn't tolerate cloud latency — real-time AR overlays, offline field inspection, privacy-sensitive medical imaging. The bounding box support is icing; ship this.

Skeptic
45/100 · skip

GPQA Diamond is one benchmark. One. Benchmark performance doesn't translate linearly to real-world task performance, especially for a merged model that hasn't been fine-tuned for instruction following or RLHF alignment. Impressive number, but I'd want to see this on coding, reasoning chains, and RAG tasks before getting excited.

45/100 · skip

450M parameters with 8-language support and benchmark-leading vision grounding sounds great until you try to fine-tune it for a domain-specific task. The LEAP platform is still invite-only and the open weights lack fine-tuning docs. Worth watching but not shipping to prod yet.

Futurist
80/100 · ship

Model merging is the dark horse of AI efficiency research. If MRI-guided DARE-TIES merging can reliably produce results like this, it suggests we're nowhere near the ceiling for extracting value from existing open-weight models. The future may involve less training and more intelligent composition.

80/100 · ship

The race to run capable VLMs on-device is the precursor to AI-native hardware. Liquid's non-Transformer architecture is showing that efficiency gains don't require the same trade-offs as quantization. This is what AI hardware of 2028 will be built around.

Creator
80/100 · ship

A capable model in the 4-5B range that can run on a MacBook M-series is exactly what solo creators need for on-device inference. If Darwin-4B-David's performance holds on creative tasks, it's a genuine local creative AI for people without cloud budgets.

80/100 · ship

On-device vision that can call functions means camera-native apps that don't phone home. Think real-time style transfer, offline image tagging, or AR creative tools that actually work on a plane. The creator tooling implications are underrated.

Weekly AI Tool Verdicts

Get the next comparison in your inbox

New AI tools ship daily. We compare them before you waste an afternoon.

Bookmarks

Loading bookmarks...

No bookmarks yet

Bookmark tools to save them for later