Skip to main content
BenchLM
Data
Leaderboard alternative
Data verified

Artificial Analysis Alternative: BenchLM Rankings

Artificial Analysis runs its own model benchmarking and publishes composite indices. BenchLM is the alternative built on published, citable results: every sourced score links to where it was reported, a verified lane excludes anything uncited, and daily-synced pricing sits next to every ranking. Start at /best/overall.

Artificial Analysis and BenchLM sit in the same aisle — independent sites helping you choose between LLMs — but they take opposite approaches to where the numbers come from. Artificial Analysis generates evaluation data by running models itself; BenchLM aggregates published benchmark results and makes their provenance the product. BenchLM even tracks the Artificial Analysis Intelligence Index as a display-only external reference, so you can see both signals side by side.

This page lays out, as of November 2026, what each approach is good for and what switching to (or adding) BenchLM gets you.

What Artificial Analysis does well

Artificial Analysis has earned its reputation as an independent benchmarking and analysis site. Running evaluations itself means consistent conditions across models and coverage that does not depend on what labs choose to publish — new releases get measured quickly, on the same footing as everything else. Its composite indices condense that work into memorable single numbers that the industry widely quotes as a market snapshot, and its analysis extends beyond quality into the operational dimensions of speed and cost. That combination — independence, consistency, and fast coverage of new models — is why BenchLM itself tracks the Artificial Analysis Intelligence Index as an external reference signal rather than ignoring it.

When you want an alternative

A composite index answers "roughly how good is this model" in one number, and that is exactly its limitation when you need to defend a decision. If a stakeholder asks why you picked a model, "it scores 80.2 on SWE-bench Verified, reported here" is a different class of answer from "it ranks highly on an index." You may want scores that trace to specific published results, an explicit split between verified and provisional data, visibility into how much sourced coverage backs each headline number, and raw data you can export and re-weight with your own priorities. If per-benchmark provenance and exportable data are what you are missing, that is the specific gap BenchLM is built to fill.

Artificial Analysis vs BenchLM

DimensionArtificial AnalysisBenchLM
Ranking methodIndependent benchmarking and analysis, summarized in composite indices such as its Intelligence IndexWeighted category scores built from published per-benchmark results, so you can always drill from the headline number to the individual scores behind it
Verification & provenanceRuns and publishes its own evaluationsSeparate verified (sourced-only, exact citations) and provisional lanes, plus 1–3 confidence dots per model showing sourced coverage
Pricing dataPublishes model pricing and performance analysisDirect provider list prices per 1M tokens, synced daily, with eight provider pricing hubs and price-vs-performance rankings
Data downloadsA Data API and MCP server, plus data export on paid plans (see the API comparison below)CSV/JSON exports of the full leaderboard on /data, plus llms.txt and llms-full.txt for LLM crawlers, free for non-commercial use (CC BY-NC 4.0)

Competitor cells stay at the level of its publicly known approach; "—" means we do not make claims about that dimension. BenchLM's cells are documented on the methodology page.

Both have a data API, and they suit different jobs

Both sites sell the data behind their pages through a REST API and a remote MCP server, so an assistant such as Claude, ChatGPT, Codex, or Cursor can query it directly. Pick Artificial Analysis when you need speed, latency, or its own image and video rankings and prices, because it measures those itself and we don't. Pick ours when you need published benchmark results with their sources, or rank and price history, at a lower self-serve price.

The free tiers differ in shape. Artificial Analysis gives 100 requests a day, which is more than our 1,000 reads a month, but only for internal use with attribution. Its paid Pro plan is $417 a month per seat with no trial, against $49 a month for Data Pro. We read their plans and docs on November 6, 2026.

DimensionArtificial Analysis Data APIBenchLM Data
MCP servermcp.artificialanalysis.ai/mcp, read-only. Sign in, or use an API key for deployed agentsdata.benchlm.ai/mcp. Sign in, or use an API key. MCP and REST share one allowance
Free tier100 requests a day, or 200 MCP calls a day when signed in. Attribution required; internal use only1,000 reads a month, 30 requests a minute, and the last 30 days of history
Self-serve paid planPro at $417 a month per seat, which covers the whole platform. No free trialData Pro at $49 a month: 25,000 reads and 3 months of history, with a 7-day trial
Top tierCommercial, quoted. The base package is 1,000 requests a day and adds provider-level data and redistribution rightsResearch at $249 a month: 100,000 reads and all retained history. Display or redistribution needs a separate license
Quality dataIts own evaluation runs and indices, such as the Intelligence IndexOur rankings, plus published benchmark results in their original units with the source and license notice on each row
Speed and latencyMeasured: output speed, time to first token, and end-to-end time, per model and per providerNot in the Data API
PricesInput, output, blended, and cache prices per million tokens, per model and providerStandard list prices per model, each linked to the provider page it was checked against, with the launch price where we have it and each recorded change since
Image, video, speech and musicIts own Arena rankings and prices for eachPublished image and video Arena results with their sources, and speech and audio prices. No media prices or music data, and no media tests of our own
HistoryPerformance over time in 7, 30, and 90-day windows on the Commercial packageRank history, price history, and dated source snapshots, limited by plan

Artificial Analysis figures come from its Data API page, API docs, and pricing page, read November 6, 2026. Our plan limits are on the Data access page.

What you get here

The main leaderboard and /best/overall rank every tracked model with two lanes: verified, built only from scores with exact citations, and provisional, which admits uncited-but-public numbers until citations are attached. Confidence dots on every model show at a glance how much sourced coverage backs its score, and each of the benchmark pages — including the Artificial Analysis Intelligence Index page itself — documents what is measured and where each score came from.

Around that core: daily-synced pricing with provider hubs (Anthropic, OpenAI, Google, DeepSeek), speed and latency snapshots, and downloadable data — CSV/JSON exports, llms.txt, and llms-full.txt, free for non-commercial use — so you can rebuild or re-weight the ranking yourself.

Questions

Is BenchLM better than Artificial Analysis?

They are different instruments. Artificial Analysis runs its own consistent evaluations and condenses them into widely cited indices; BenchLM aggregates published benchmark results and makes every sourced score auditable. If you want fast, independent measurement of new releases, Artificial Analysis does that well. If you want per-benchmark provenance, a verified-only lane, and exportable data, that is BenchLM. Using both is reasonable — BenchLM even displays the AA Intelligence Index as a reference.

Does BenchLM use Artificial Analysis data?

BenchLM tracks the Artificial Analysis Intelligence Index as a display-only external reference — useful as a market snapshot, but it is not a weighted input to BenchLM's own scores, which are built from published per-benchmark results as described in the methodology.

Why do the two sites rank models differently?

Different inputs and different weights. Artificial Analysis scores come from its own evaluation runs combined into composite indices; BenchLM scores come from published results across dozens of benchmarks, with agentic, coding, and reasoning work weighted most heavily. A model with strong lab-reported benchmark results but a middling index score (or the reverse) will land in different places — comparing the two is often more informative than either alone.

Does Artificial Analysis have an API?

Yes. Its Data API and MCP server cover its indices, prices, speed, and media rankings. The free tier allows 100 requests a day for internal use with attribution, and Pro costs $417 a month per seat. Our Data API starts free and costs $49 a month on Pro. The comparison above lists the differences.

Is BenchLM free and open?

Yes. The rankings, benchmark pages, and pricing hubs are free, and the underlying dataset is published for reuse via CSV/JSON exports, llms.txt, and llms-full.txt.

Leaderboard updates

Get notified when new benchmark results reshuffle the verified rankings.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.