Research Hub
BenchGecko Research connects model rankings, benchmark evidence, API prices, compute constraints, and market attention in one readable view.
The evidence behind the rankings
BenchGecko separates capability, price, infrastructure, and attention signals so readers can see which claims are benchmarked, sourced, and comparable.
- Model evaluation entry points
- Benchmark and source discovery
- Compute and market context
Research library
Papers · notes · methods · benchmark sources · datasets · API access
Methods first. Papers when they are ready.
BenchGecko lists live methods, datasets, benchmark records, and research notes now. Formal papers will be published only after authorship, review, version dates, stable URLs, and citation text are in place.
Benchmark Saturation Watch
Tracks where public evaluations still separate frontier models and where scores are clustering too tightly to explain real deployment choices.
API Pricing Compression Monitor
Reads the current spread of input and output prices across every priced model and provider, so model quality can be read beside cost pressure.
Compute Pressure Index Notes
Explains how chips, memory, foundries, systems and energy feed the AI Compute Demand Index, using the curated infrastructure dataset and its as-of date.
Search papers, notes, methods, and benchmark sources
BenchGecko model ranking methodology
How BenchGecko normalizes models, providers, benchmark scores, pricing fields, caveats, and comparison logic.
AI compute pressure methodology
How chips, foundries, memory, systems, power, capex, and regional readiness are separated into compute signals.
Mindshare methodology
How BenchGecko measures attention from Hacker News mentions, Wikipedia page views and GitHub stars, kept separate from capability scores.
Model pricing dataset
Normalized model price records for input, output, cache, reasoning, and API access comparisons across providers.
BenchGecko research API
Machine readable access paths for researchers and builders who want to use BenchGecko data in their own systems.
Frontier model evaluation notes
A current view of frontier and open model movement across normalized benchmark records and provider metadata.
AI compute pressure note
Current infrastructure pressure across chips, memory, systems, energy, capex, regions, and supply constraints.
BenchGecko papers index
Formal BenchGecko papers and downloadable research reports will live here once authored, reviewed, and versioned.
Benchmark Saturation Watch
Tracks where public evaluations still separate frontier models and where scores are clustering too tightly to explain real deployment choices. 1,235 models · 158 tracked benchmarks.
API Pricing Compression Monitor
Reads the current spread of input and output prices across every priced model and provider, so model quality can be read beside cost pressure. 273 providers · 595 priced models.
Compute Pressure Index Notes
Explains how chips, memory, foundries, systems and energy feed the AI Compute Demand Index, using the curated infrastructure dataset and its as-of date.
Aider — Code Editing
Code editing benchmark from the Aider project. Measures ability to apply targeted code changes while maintaining correctness and style.
BenchGecko publications and research notes
Formal papers later · live methods and notes now
Formal papers
Formal BenchGecko papers will need authorship, review, version dates, citation text, stable URLs, and downloadable files before publication.
Research notes
Research notes can track benchmark saturation, pricing compression, compute pressure, model movement, and mindshare shifts using live BenchGecko data.
Methods and datasets
Public methodology, pricing, benchmark, compute, and API access pages are indexed in the searchable library above.
Research overview
Models · Benchmarks · Pricing · Compute · Economy · Mindshare
Model evaluation
Frontier and open model rankings built from normalized benchmark coverage, model metadata, provider mapping, and capability summaries.
Benchmark evidence
Coverage across coding · knowledge · agentic · reasoning and 7 more categories, with source links where available.
Pricing intelligence
Input, output, cached, and access pricing normalized into comparison pages for builders choosing APIs.
Compute infrastructure
Significant bottlenecks in AI compute supply chain. BenchGecko tracks chips, foundries, memory, systems, regions, energy, capex, and AI power pressure.
AI economy
Company pages, funding, valuation, revenue, capex, and infrastructure pressure around the AI economy.
Mindshare signals
Daily attention around models, companies, agents, people and topics: Hacker News mentions, Wikipedia page views and GitHub stars, with weekly movement.
Research map
Models · benchmarks · pricing · compute · economy · mindshare
Start with the question. Follow the evidence.
Use this page as the front door for BenchGecko research. Start with a model, benchmark, price, compute signal, company, or mindshare trend, then jump to the page with the numbers behind it.
How BenchGecko normalizes model, provider, pricing, and benchmark data.
How infrastructure pressure and regional readiness are calculated.
How attention signals stay separate from capability scores.
Machine readable paths for builders and researchers.
AI model rankings
Covers rankings · model cards · score coverage. Follow the numbers into the live BenchGecko data.
LLM benchmark research
Covers benchmark records · papers · eval repositories. Follow the numbers into the live BenchGecko data.
AI pricing data
Covers token costs · provider pages · pricing notes. Follow the numbers into the live BenchGecko data.
AI compute infrastructure
Covers chips · datacenters · memory · energy. Follow the numbers into the live BenchGecko data.
AI company economy
Covers capex · funding · company data. Follow the numbers into the live BenchGecko data.
AI attention signals
Covers Hacker News · Wikipedia · GitHub. Follow the numbers into the live BenchGecko data.
Explore the evidence
Jump from the research index into model, benchmark, provider, pricing, and methodology views.
Model ranking snapshot
Current BenchGecko rankings · top models by BenchGecko score (normalized across benchmarks)
Top ranked models
What the ranking can explain
This view explains what supports the public rankings: what is measured, what is priced, what has source context, and where the gaps remain.
Benchmark source map
Source URLs attached to benchmark records where available
Compute index
AI compute demand index · regional readiness · capex context · Updated Apr 14, 2026
Methods and citation path
How the research hub connects back to source pages
Models, providers, benchmarks, prices, compute, and mindshare records.
Map records into comparable model, provider, benchmark, and topic layers.
Expose rankings, coverage, pressure, and attention as separate signals.
Link each claim to the specific BenchGecko page that supports it.
What does BenchGecko Research cover?
It covers model rankings, LLM benchmark evidence, API pricing, compute infrastructure, company data, and mindshare signals. Each section links to the relevant BenchGecko page for the details.
How fresh is the data?
Models, prices and benchmark scores are refreshed every day by automated jobs, and Mindshare signals (Hacker News, Wikipedia, GitHub) are collected daily. Compute, memory and energy figures come from a curated dataset with its own as-of date. Each section shows the date of the data it uses.
Why does BenchGecko track benchmark sources?
Benchmark pages, papers, model releases, and open projects are the evidence behind capability claims. BenchGecko keeps that evidence close to the model, price, provider, and compute data it affects.
How should researchers cite BenchGecko?
Link to the specific model, benchmark, pricing, compute, or methodology page when possible. Use this page when citing the full research index rather than one record.