ModelSpec is an analysis of alternatives for AI models: it decides which models fit a job from your requirements, sourced benchmarks and real cost, and shows its work.

Long documents, self-hosted

Find a commercially usable open-weights model with a long context that you can run yourself.

Worked example as of 2026-09-23, from the public data image.

Open weights with a long context; commercial licence.

Requirements

model.class = text-generator: The result must be generated as text.
model.weights_openness = open_weights: Self-hosting requires downloadable weights; add a model.fits_hardware Must for the device you own.
licence.commercial_use in {permitted, permitted_with_conditions}: The licence must allow commercial use, including use with stated conditions.
model.context_window >= 128000: The model must accept a long document in one request.

Preferences

writing: Prefer clear, high-quality long-form writing among self-hostable models.
model.context_window: Prefer the longer context among self-hostable models.

Applied template spec

spec_version

where

model.class = text-generator
model.weights_openness = open_weights
licence.commercial_use in {permitted, permitted_with_conditions}
model.context_window >= 128000

optimize

weights

writing: 0.6
model.context_window: 0.4

Worked answer

Decision status: partial.

No recommendation: the public image has no answer with enough evidence.

This is a partial answer. Unknown facts or missing objective evidence can leave eligible candidates unranked.

probability bands: a model joins the leader when P(its score >= the leader's) >= 0.25 under the score posteriors; a weighted capability with an 80% interval wider than 2.8 is not enough evidence

Score and score interval use the objective's dimensionless normalised units. Cost per task is for the engine-selected offering and the applied spec's task size. When task size is omitted, the engine uses its documented default. P(best) is the engine probability. Unavailable means the engine did not return a value. Thin evidence is undetermined and is not a recommendation.

Thin evidence: undetermined

Model	Score	Score interval	P(best)	Cost per task, USD	Capability estimates and intervals
`zhipu/glm-5-3`	1.0	[-0.536561, 2.536561]	0.29296875	0.0736	writing: 0.009738247; interval [-1.489401109, 1.508877604]; benchmarks 3; direct benchmarks 0
`moonshot/kimi-k3`	0.936072	[-0.600488, 2.472633]	0.28125	Unavailable	writing: -0.052632519; interval [-1.551771876, 1.446506837]; benchmarks 3; direct benchmarks 0
`zhipu/glm-5-2`	0.693397	[-0.843164, 2.229957]	0.19140625	0.0736	writing: -0.265292672; interval [-1.764432028, 1.233846685]; benchmarks 3; direct benchmarks 0
`deepseek/deepseek-v4-pro`	0.548556	[-0.988004, 2.085117]	0.14453125	0.06864	writing: -0.406605901; interval [-1.905745257, 1.092533456]; benchmarks 3; direct benchmarks 0
`moonshot/kimi-k2-6`	0.0	[-1.536561, 1.536561]	0.08984375	Unavailable	writing: -0.575649419; interval [-2.074788775, 0.923489938]; benchmarks 3; direct benchmarks 0

Unknown requirements

Some candidates may qualify. Missing facts:

model.weights_openness
writing

Engine warnings

not_separable
proxy_evidence_only

Decision dec_8ba9eea1c47cefde688ffa6e on public snapshot snap_05a963d2477ac2fb.

What this template teaches

Self-hosting narrows the field, so the context floor is lower and longer context becomes a Prefer. Lab jurisdiction is a buyer-set Must (origin.lab_jurisdiction in {…}); an unverified jurisdiction does not pass, so set the countries on the facet board. The same condition works as an inventory profile rule.

The live board uses current data. Its answer can change as evidence changes.

Apply this template on Decide

Human lookup is free on Decide. CLI, API, and MCP access requires a key. Access and pricing.

Use-case examples · How we decide