How the Indices tab works

Why a separate Indices tab exists

The Overview tab shows the state of the market. The Signals tab reveals structural properties of it. The Indices tab does something different: it establishes numerical benchmarks that let observers say, next year, "this went up" or "this went down" - with a specific rate and against a specific baseline.

Overview and Signals answer "what does the market look like now." Indices answers "how has it been moving, and how fast." That distinction is why the tab is separate and why it sits between Overview (state) and Signals (structure).

Why four indices, and why publish two at zero

A maturing field eventually acquires the second layer that any serious discipline eventually needs: agreed-upon numerical measures that let observers make year-over-year comparisons. We are proposing four for this niche. Two are measurable today; two hold at zero because no confirming public data yet exists.

Publishing a metric at zero is deliberate. If we waited until Weaponization first became nonzero to establish the index, we would lose the ability to say, in retrospect, "here is the exact date the number moved for the first time." Zero is not a null value. Zero is the reading, and it is worth having on record so that the transition to nonzero is measurable.

The four indices at a glance

Freedom Velocity Index

The average number of days from a base model family's first appearance to its first abliteration. Faster reaction times mean the family is a popular target; slower reaction times mean it took a while to catch attention.

Current reading: 65.8 days across 19 tracked families. The fastest families (Qwen, GLM, DeepSeek) reach zero or single-digit days. Older families (Gemma, Mistral) show weeks-to-months lag. If tooling keeps getting faster (Heretic, hosted services), we expect this index to trend toward zero.

Full methodology in the three-indices article.

Weaponization Index

The share of indexed abliterated models publicly named in a documented case of real-world harm - a court case, a formal regulatory action, a peer-reviewed security disclosure, or a first-party AI-lab threat report that names the specific model.

Current reading: 0.0%. Zero verified incidents as of the latest review of research firms (Palo Alto Unit 42, Cato Networks, SlashNext, Abnormal Security), news outlets, AI-lab threat reports (Anthropic, OpenAI, Google GTIG), and Hugging Face Trust and Safety records. The famous "criminal LLMs" (WormGPT, FraudGPT, GhostGPT) are wrappers or fine-tunes over commercial models, not abliterated models in the technical sense.

Research demonstrations show potential harm but do not count. This index counts realized harm.

Freedom-to-Weapon Latency

The average number of days between the first abliteration of a family and the first documented misuse of that family.

Current reading: N/A. Not measurable while the Weaponization Index is zero. Once Weaponization moves, this becomes the more useful of the two: it shows the gap between a technique becoming available and its first documented misuse. Short latency would mean attackers actively watch for new abliterations; long latency would mean abliteration risk is largely theoretical for the near term.

Disenchantment Index (DAI)

The share of each base model's download market that has flowed into abliterated descendants at any depth in the fork tree. Aggregated across all tracked bases as a weighted market-share value.

Current reading: 8.93% weighted across 478 bases. Up from 0.15% in 2024-10 - a 58x shift over 24 months.

Full mathematical treatment plus etymology (why the name references Weber) in the Disenchantment Index article. The detail page holds the full ranked table with sparklines and click-to-expand charts.

Below the four indices, on the same tab

After the four-index panel, the Indices tab includes a Disenchantment detail block: the flagship DAI number bigger, plus two 24-month charts (the DAI level over time, and its rolling growth rate), plus a live top-10 table of the most-disenchanted bases. This puts the most active of the four indices front and center without displacing the other three.

How the two measurable indices tell different halves of one story

Freedom Velocity is a supply-side speed metric: how quickly the community responds to a new open-weight release. Small numbers mean the ecosystem is agile - within days of a new base landing, an abliteration exists.

DAI is a demand-side concentration metric: what share of the audience actually chose the abliterated version once it existed. Small numbers mean the audience mostly stayed with the base; large numbers mean the abliterated variants captured meaningful market share.

Together they describe the pipeline: how fast the response happens, and how large it becomes. A niche can have fast Freedom Velocity but low DAI (community produces quickly, but nobody wants what they produce), or slow Freedom Velocity but high DAI (production is slow, but every release finds an audience). Our current readings show fast production and rising DAI - both halves are moving.

When Weaponization and Latency will first move

Weaponization moves the first time a specific abliterated model is publicly named in a documented case of realized harm: a court filing, a regulatory action, a peer-reviewed security disclosure, or a first-party AI-lab threat report. The bar is high on purpose - low-bar inclusion (research demonstrations, near-misses, unattributed suspicion) would produce a noisy indicator that is not useful for year-over-year comparison.

Once Weaponization moves, Freedom-to-Weapon Latency becomes measurable in the same instant: the gap between the first abliteration of the incident's family and the incident date. The two indices are coupled - one gates the other.

Our monitoring includes weekly deep-research passes over security research firms, news outlets, AI-lab threat reports, and Hugging Face Trust and Safety records. The moment a confirming incident lands, the index will move on the next site rebuild.

What the Indices tab is not

Not safety benchmarks. WMDP and HarmBench probe model behavior with curated prompts. We do not run inference. We count published events and market flows.

Not predictions. An index at zero today does not mean the number stays zero tomorrow. It means the number is zero today.

Not value judgments. Freedom Velocity going up is neither good nor bad without context. Weaponization moving from zero to one is a fact, not an accusation. DAI rising is a market signal, not a moral score.

Where to read more

• Three frontier indices - full methodology for Freedom Velocity, Weaponization, Freedom-to-Weapon Latency
• Disenchantment Index (DAI) - full mathematical and etymological treatment
• /market/disenchantment - the ranked table of every base with abliterated descendants
• How market metrics work - the master guide to all three tabs
• How the Signals tab works - companion article for the sibling tab