Review Methodology

Hands-On Reviews

When we have direct access to a product, we may test representative tasks such as installation, setup, response quality, coding, long-context use, tool calling, image understanding, latency, pricing behavior, compatibility, or workflow reliability. We aim to describe the environment, model or software version, important settings, test prompts or steps, and any limitation that could materially affect the result.

We do not describe an article as hands-on if we did not directly use the product for the claims being made.

Source-Based Reviews and Comparisons

Some new products cannot be tested directly at publication time. In those cases, we may publish a source-based review using official documentation, release notes, model cards, repositories, pricing pages, research material, and reputable independent measurements. The article should make that method clear and should not present vendor benchmark results as if they were our own tests.

Benchmarks

Benchmark scores are useful only when their source and conditions are understood. We identify whether a result comes from the developer, an independent evaluator, or our own test. Where practical, we compare more than one source and explain differences in reasoning settings, model variants, prompts, harnesses, context limits, or infrastructure that may affect the result.

Screenshots and Evidence

We use screenshots, interface captures, benchmark figures, command output, tables, and other evidence when they help readers verify a point. Images should be attributed when they originate from an official or third-party source. Original screenshots should avoid exposing API keys, personal information, account identifiers, tokens, or other secrets.

Pricing and Availability

AI pricing and model availability can change quickly. We record the date of checks when a price, promotion, provider route, or availability claim is time-sensitive. Readers should verify live pricing before committing significant spend.

What We Evaluate

• Capability relative to the product’s stated purpose.
• Accuracy and limitations observed or documented.
• Ease of installation and setup.
• Speed, reliability, and workflow fit when measurable.
• Pricing and value relative to relevant alternatives.
• Privacy, security, permissions, and deployment considerations.
• Open-source status, licensing, and self-hosting options where relevant.
• Whether marketing claims are supported by evidence.

Corrections

If new evidence changes a conclusion, we may update the article. Material factual errors should be corrected promptly. To report an issue, email contact@aitechledger.com and include the article URL and a reliable supporting source.

For broader standards, see our Editorial Policy.