We test technology with Chilean data and documents before we recommend it.
At ITLAB we test AI models that run on your own hardware, without sending data to external services. We evaluate them with a test battery we designed: receipts, invoices, RUT numbers, VAT amounts, dates, and Chilean Spanish text.
We want to answer a practical question: what can a local model solve today, on what hardware, and with what margin of error?
Categories
What we publish covers what we test in the lab and what we apply every day with our clients.
Articles
Analysis, tests, guides and news from the lab. Each one shows its publication date.
Page 1 of 2
Swift 27B vs. two Flash Next models on the IT Lab Bench
Three local AI models, 153 real-work cases on one 16 GiB GPU: results, speed, memory and IT Lab's guidance for business use.
Read article Cybersecurity and regulation22 / 10 / 2026 · 5 minPHP fixes two critical vulnerabilities: update now
PHP released patches for two critical remote code execution flaws (CVSS v4.0 9.2). Fixed versions, who is affected and what to do right away.
Read news From the lab21 / 10 / 2026 · 4 minLocal AI for a Chilean company: hardware and limits (2/2)
What hardware local AI needs in a Chilean company, what IT Lab does not promise, frequently asked questions, limitations and measurement sheet.
Read guide From the lab21 / 10 / 2026 · 5 minLocal AI for a Chilean company: what to automate (1/2)
Which office and development tasks a local model can automate today, which need human review, and what hardware it takes.
Read guide From the lab21 / 10 / 2026 · 7 minSMB hardware for local AI with 16 GB
What fits on a 16 GiB graphics card, and why supporting a long context is not the same as working at good speed with…
Read article From the lab20 / 10 / 2026 · 7 minConfiguring a local model matters as much as choosing it
In the IT Lab Bench, the same model changes in speed and in outcome depending on its configuration, so it is documented per model…
Read article From the lab14 / 10 / 2026 · 7 minHard programming in real repositories with local AI
Repositories of 10 to 15 files with two interacting faults: several local models pass 9 or 10 of 10, and generation speed does not…
Read article From the lab07 / 10 / 2026 · 5 minLocal AI for Chilean receipts and tax documents
IT Lab applies its own criteria to reading Chilean tax documents with local AI, so company data never leaves the organization.
Read article From the lab06 / 10 / 2026 · 7 minLocal AI reasoning on Chilean business cases
40 Chilean small-business cases scored with automatic controls: no model passes them all, and extra reasoning fixes only part of the failures.
Read article From the lab01 / 10 / 2026 · 7 minLocal AI for code: completing a base vs creating
On an existing base with marked gaps, three local models meet the criterion; creating from scratch, none. Paid models do.
Read article From the lab30 / 09 / 2026 · 7 minAgentic work with local AI: finishing the job
Three episodes with tools and actions: 5 of 14 models meet the criterion, the best is local, and the characteristic failure is not closing…
Read article From the lab29 / 09 / 2026 · 7 minVision: images and scanned documents
Five of six eligible models score 97 or higher in vision and two local ones reach 97. The limit is tiny or degraded text,…
Read article From the lab28 / 09 / 2026 · 7 minDocuments with local AI: extract yes, count no
In the IT Lab Bench, 12 of 14 models extract all 183 cells of a real document with no error; counting over what they…
Read article From the lab22 / 09 / 2026 · 7 minWhy AI benchmarks miss real work
High scores on HumanEval and BigCodeBench did not rank local models by usefulness; IT Lab therefore measures tasks completed without human intervention.
Read article From the lab17 / 09 / 2026 · 7 minHow IT Lab evaluates a local AI model
The IT Lab Bench measures real work with fixed automatic judges, gives each model its best configuration and publishes what it measures and why.
Read article Continuity and IT support14 / 09 / 2026 · 5 minWebsite Launch Checklist for a Business Site or Portal
What to settle before going live: subdomain, DNS, hosting, keeping the page unindexed until legal approval, and email sent from your own domain.
Read news Personal data protection01 / 09 / 2026 · 5 minLocal AI: Using Artificial Intelligence Without Sharing Data
How to apply artificial intelligence to internal information without data leaving the organisation: local models, limits and evaluation criteria.
Read news Personal data protection18 / 08 / 2026 · 5 minA Chilean IT Provider for Clients in the United States
How a Chilean IT provider works with clients abroad: servers in their territory, their requirements, a counterpart with equal access, clean exit.
Read news Continuity and IT support05 / 08 / 2026 · 5 minWho Owns Your Data: What to Require From a Software Vendor
Data and licences in the client's name, testing on a copy and exit without lock-in: IT governance criteria for choosing and changing a software…
Read news Cybersecurity and regulation22 / 07 / 2026 · 5 minVirtual Data Room for Investors: What to Require
How to design a data room or private investor portal: individual accounts, separate roles, access logs and protected downloads, set up before launch.
Read news Continuity and IT support09 / 07 / 2026 · 4 minWebsite Email with Cloud Mail: DNS and Permissions
A tidy guide to set up your website's outgoing email with your corporate cloud mail platform: DNS and least-privilege permissions. Ask IT Lab for…
Read news Continuity and IT support26 / 06 / 2026 · 4 minSecure Customer Portal: Isolate the Operational DB
How to design a customer portal that protects the operational database through a read-only replica and an API. Ask IT Lab about secure portals.
Read news Cybersecurity and regulation12 / 06 / 2026 · 5 minAI Risk Analysis in Mining: Support for Human Judgement
How to use artificial intelligence as support for risk analysis in mining operations, with clear limits and human oversight. Request guidance from IT Lab.
Read news Personal data protection01 / 06 / 2026 · 4 minDeclaring What Does Not Apply: AI, APIs, Dependencies
Why justifying in writing the controls that do not apply to your company strengthens a security audit or assessment. Ask IT Lab for guidance.
Read newsTechnical reports
Reference pieces with data: methodology, measurements and IT Lab Bench results, in reading order.
Technical report 1: office capabilities of local AI
Document reading, vision, agentic work and reasoning on Chilean business cases: full results from the IT Lab Bench, edition 1.
Read report From the lab15 / 10 / 2026 · 18 minTechnical report 2: Local AI for writing code
IT Lab Bench results on code: completing an existing base, creating from scratch and realistic programming, with a staged judge.
Read report From the lab21 / 10 / 2026 · 18 minTechnical report 3: Running local AI on a 16 GB GPU
Usable memory, engine configuration, real context, the cost of reasoning and sizing, as measured by IT Lab on a 16 GiB GPU.
Read report From the lab21 / 10 / 2026 · 8 minTechnical report 4: IT Lab Bench methodology (1/2)
The IT Lab Bench method: test families, deterministic staged judge, threshold set in advance, best opportunity per model and instrument quality controls.
Read report From the lab21 / 10 / 2026 · 9 minTechnical report 4 (2/2): results and conclusions
Results, discussion, implications, limitations and conclusions of the IT Lab Bench methodology; second part of technical report 4.
Read report From the lab21 / 10 / 2026 · 6 minIT Lab Bench results, edition 1
Per-model results by test family from the IT Lab Bench, edition 1: pass criterion and time, with a summary matrix.
View resultsWant to bring this to your operations?
Let's talk about the state of your business and the priorities worth tackling. Each part of the series ends with a link to the next and with the option to talk with us.
Let's talk about your business