VC
Value Add VC
⚡HomePulse⚡Helpful Apps📝Blog
← Value Add PulseAI

Liquid AI's Tiny LFM2.5-230M Beats Models 4x Its Size and Runs Anywhere

Liquid AI released LFM2.5-230M, its smallest model yet, which the company says outperforms models four times its size at data extraction while being small enough to run 'anywhere' -- including phones, laptops and edge devices. The release advances the counter-narrative to ever-larger models: that efficient, specialized small models can win on the tasks enterprises actually run at scale.

LFM2.5-230M
Model
230M parameters
Size
Beats 4x-larger models
Claim
Data extraction
Task
Phones, laptops, edge
Deploys
TC
Trace Cohen
Early-stage VC & angel · Founder, New York Venture Partners
June 25, 2026
2 min read
KEY TAKEAWAYS FOR VCs & FOUNDERS
1

Small models that beat bigger ones attack the cost and latency of real enterprise AI workloads

2

On-device capability sidesteps cloud costs, privacy concerns and the GPU bottleneck entirely

3

It strengthens the case that the future of AI is many specialized small models, not one giant one

4

Liquid AI's efficiency pitch is a differentiated lane away from the frontier-scale arms race

TC
The VC Read · Trace's TakeTrace Cohen

The frontier gets the headlines, but the quiet money is in efficiency -- and a 230M model that beats systems 4x its size on real extraction tasks is exactly the kind of unglamorous win enterprises pay for. On-device is the underrated part: no cloud bill, data stays local, no GPU queue, and -- timely this week -- no government gate on access. The thesis that the future is many small specialized models, not one giant one, keeps getting more credible. Watch for independent benchmarks; small-model claims are easy to make and the real test is whether the accuracy holds outside the lab's own evals.

🤖 AI Landscape →

Liquid AI has released LFM2.5-230M, the smallest model in its lineup, which the company says outperforms models four times its size at data-extraction tasks while being compact enough to run virtually anywhere -- from smartphones and laptops to constrained edge devices. At just 230 million parameters, it is a fraction of the size of the frontier models that dominate headlines.

The release is a pointed argument in one of AI's most important debates: whether bigger is always better. For many high-volume enterprise tasks -- extracting fields from documents, classifying records, parsing structured data -- a small, specialized model that runs locally can deliver comparable or better accuracy at a tiny fraction of the cost and latency of calling a giant cloud model. Liquid AI, an MIT-spinout known for its efficiency-focused architectures, is betting that this practical reality wins real deployments.

“The release is a pointed argument in one of AI's most important debates: whether bigger is always better.”

The 'runs anywhere' claim matters strategically. On-device inference eliminates per-call cloud costs, keeps sensitive data local for privacy and compliance, and sidesteps the GPU-supply bottleneck that has throttled cloud AI. For enterprises wary of sending data to external models -- a concern sharpened by this week's news that the US government is now vetting access to frontier models -- a capable model that runs on their own hardware is increasingly attractive.

The competitive landscape for small, efficient models is heating up: Microsoft's Phi family, Google's Gemma, Meta's smaller Llama variants, Mistral's compact models and a growing field of on-device specialists all compete on the performance-per-parameter frontier. Liquid AI differentiates on its novel architecture and an explicit focus on efficiency rather than raw scale, a lane that grows more valuable as the cost of running large models becomes a board-level concern.

The bear case is that benchmark claims need independent validation, small models hit real ceilings on complex reasoning, and the giants can ship their own efficient variants. What to watch: third-party verification of the data-extraction claims, enterprise adoption for on-device workloads, and whether the market truly bifurcates into frontier models for hard reasoning and small models for everything else.

ShareXLinkedInEmail

Originally reported by VentureBeat. Analysis and editorial commentary by Value Add Pulse.

← Back to Pulse

Markets Now

live
SPCX▲+0.73%
$233.10
CBRS▲+0.66%
$259.80
SPY▲+0.07%
5,952.40
QQQ▲+0.12%
20,061.30
NVDA▼-0.98%
$152.10
MSFT▲+0.29%
$482.60
GOOGL▲+0.34%
$209.10
META▲+0.23%
$655.40

Read Next

AIGPT-5.6 (Sol/Terra/Luna)

OpenAI Launches GPT-5.6 -- But the Government Decides Who Gets to Use It

OpenAI unveiled its most capable models yet -- GPT-5.6 Sol, Terra and Luna -- but at the Trump administration's request limited the initial release to a small group of trusted partners whose identities are shared with the government. OpenAI complied while publicly objecting, warning that 'this kind of government access process should not become the long-term default.' The move mirrors restrictions placed on Anthropic's frontier models and marks the first time the US government is effectively vetting who can access America's leading AI systems.

AI

OpenAI's Updated GPT-5.5 Instant Gets Better at Shopping and Complex Constraints

OpenAI shipped an updated GPT-5.5 Instant that is better at shopping, handling complex constraints, and understanding user intent -- and it's already live in the API. The release sharpens OpenAI's fast, low-latency tier for agentic and commerce use cases, even as the company's more powerful new GPT-5.6 family sits behind a government-vetted gate.

AI

A New Agentic Memory Framework Uses 118K Tokens Per Query -- LangMem Burns 3.26M

A newly detailed agentic memory framework reportedly answers queries using about 118,000 tokens, versus roughly 3.26 million for the widely used LangMem approach -- a ~28x efficiency gain. As AI agents take on longer, multi-step tasks, how they store and retrieve memory is becoming a decisive factor in both cost and capability.

@Trace_Cohen·t@nyvp.com