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.
Coding, self-hosted
Find a commercially usable open-weights coding model that you can run on your own hardware.
Worked example as of 2026-09-23, from the public data image.
Open weights you run yourself; commercial licence.
Requirements
model.class = text-generator: Coding work needs a model that generates text and code.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.
Preferences
software_engineering: Rank only by the software-engineering capability estimate among self-hostable models.
Applied template spec
spec_version- 1
where- model.class = text-generator
- model.weights_openness = open_weights
- licence.commercial_use in {permitted, permitted_with_conditions}
optimizeweightssoftware_engineering- 1.0
Worked answer
Decision status: partial.
Tied on the evidence
moonshot/kimi-k3zhipu/glm-5-3
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
Tie-breakers describe members of the tie. They do not establish a winner.
- Cheapest tie-breaker: zhipu/glm-5-3
- Most independently measured tie-breaker: moonshot/kimi-k3
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.
Best band
| Model | Score | Score interval | P(best) | Cost per task, USD | Capability estimates and intervals |
|---|---|---|---|---|---|
moonshot/kimi-k3 | 1.0 | [0.570896, 1.429104] | 0.35546875 | Unavailable |
|
zhipu/glm-5-3 | 0.88943 | [0.46957, 1.30929] | 0.2421875 | 0.0736 |
|
Rest band
| Model | Score | Score interval | P(best) | Cost per task, USD | Capability estimates and intervals |
|---|---|---|---|---|---|
zhipu/glm-5-2 | 0.367118 | [-0.004893, 0.739129] | 0.0 | 0.0736 |
|
moonshot/kimi-k2-6 | 0.355077 | [-0.241414, 0.951568] | 0.0703125 | Unavailable |
|
deepseek/deepseek-v4-pro | 0.0 | [-0.443998, 0.443998] | 0.0 | 0.06864 |
|
Thin evidence: undetermined
| Model | Score | Score interval | P(best) | Cost per task, USD | Capability estimates and intervals |
|---|---|---|---|---|---|
qwen/qwen3-8-flash-next | 0.843058 | [-0.387736, 2.073851] | 0.33203125 | Unavailable |
|
Unknown requirements
Some candidates may qualify. Missing facts:
model.weights_openness
Engine warnings
- not_separable
Decision dec_c4fed650835ff3d8850df9b4 on public snapshot snap_05a963d2477ac2fb.
What this template teaches
Privacy is a Must on the weights and the licence; the canvas shows the model size you would have to host. 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.
Human lookup is free on Decide. CLI, API, and MCP access requires a key. Access and pricing.