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, best available

Find the active coding model with the strongest software-engineering evidence, whatever it costs.

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

Strongest coding evidence, cost no object.

Requirements

model.class = text-generator: Coding work needs a model that generates text and code.
model.lifecycle = active: An active model is still supported by its lab or provider.

Preferences

software_engineering: Rank only by the software-engineering capability estimate.

Applied template spec

spec_version

where

model.class = text-generator
model.lifecycle = active

optimize

weights

software_engineering: 1.0

Worked answer

Decision status: partial.

Tied on the evidence

anthropic/claude-opus-5-5
anthropic/claude-sonnet-5-5
openai/gpt-6-astra
anthropic/claude-opus-4-7

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: anthropic/claude-sonnet-5-5
Most independently measured tie-breaker: anthropic/claude-opus-4-7

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
`anthropic/claude-opus-5-5`	1.0	[0.661373, 1.338627]	0.2109375	0.24	software_engineering: 1.300758259; interval [0.373384859, 2.22813166]; benchmarks 4; direct benchmarks 2
`anthropic/claude-sonnet-5-5`	0.892297	[0.538219, 1.246374]	0.1015625	0.12	software_engineering: 1.005798442; interval [0.036112567, 1.975484317]; benchmarks 3; direct benchmarks 2
`openai/gpt-6-astra`	0.788696	[0.546009, 1.031383]	0.0546875	0.6	software_engineering: 0.722074908; interval [0.05744525, 1.386704566]; benchmarks 7; direct benchmarks 2
`anthropic/claude-opus-4-7`	0.784952	[0.517382, 1.052522]	0.03515625	0.3	software_engineering: 0.711822213; interval [-0.020952777, 1.444597204]; benchmarks 5; direct benchmarks 2

Rest band

Model	Score	Score interval	P(best)	Cost per task, USD	Capability estimates and intervals
`anthropic/claude-fable-5`	0.77513	[0.526207, 1.024053]	0.03515625	0.6	software_engineering: 0.68492301; interval [0.003215236, 1.366630785]; benchmarks 6; direct benchmarks 2
`anthropic/claude-fable-5-1`	0.761648	[0.518942, 1.004355]	0.01953125	0.6	software_engineering: 0.648000878; interval [-0.016682066, 1.312683821]; benchmarks 7; direct benchmarks 2
`anthropic/claude-opus-5`	0.695731	[0.483883, 0.907578]	0.00390625	0.3	software_engineering: 0.467477024; interval [-0.11269439, 1.047648438]; benchmarks 7; direct benchmarks 3
`moonshot/kimi-k3`	0.657991	[0.404227, 0.911755]	0.0	Unavailable	software_engineering: 0.364121531; interval [-0.330844073, 1.059087136]; benchmarks 5; direct benchmarks 2
`zhipu/glm-5-3`	0.592601	[0.344304, 0.840899]	0.0	0.0736	software_engineering: 0.185044748; interval [-0.494949533, 0.86503903]; benchmarks 6; direct benchmarks 2
`openai/gpt-5-6-sol`	0.561534	[0.345877, 0.777191]	0.0	0.24	software_engineering: 0.099962916; interval [-0.490641673, 0.690567505]; benchmarks 6; direct benchmarks 3
`google/gemini-3-7-flash`	0.561464	[0.313157, 0.809772]	0.0078125	0.045	software_engineering: 0.099771765; interval [-0.580250011, 0.779793541]; benchmarks 6; direct benchmarks 2
`google/gemini-3-8-flash`	0.480862	[0.272647, 0.689077]	0.0	0.045	software_engineering: -0.120968438; interval [-0.69119135, 0.449254473]; benchmarks 8; direct benchmarks 3
`xai/grok-4-7`	0.475687	[0.137033, 0.814342]	0.0	0.104	software_engineering: -0.135139749; interval [-1.062588258, 0.792308759]; benchmarks 3; direct benchmarks 1
`openai/gpt-5-4`	0.464587	[0.242344, 0.686831]	0.0	0.16	software_engineering: -0.165538069; interval [-0.774180651, 0.443104514]; benchmarks 9; direct benchmarks 3
`anthropic/claude-opus-4-6`	0.427499	[0.20457, 0.650429]	0.0	0.3	software_engineering: -0.267108807; interval [-0.877629968, 0.343412354]; benchmarks 9; direct benchmarks 3
`meta/muse-spark-1-3`	0.36989	[0.04466, 0.69512]	0.00390625	0.067	software_engineering: -0.424879214; interval [-1.31556301, 0.465804581]; benchmarks 3; direct benchmarks 1
`google/gemini-3-5-flash`	0.350481	[0.08644, 0.614521]	0.0	0.096	software_engineering: -0.478034522; interval [-1.201143536, 0.245074492]; benchmarks 5; direct benchmarks 2
`zhipu/glm-5-2`	0.283716	[0.063716, 0.503716]	0.0	0.0736	software_engineering: -0.660878011; interval [-1.26337721, -0.058378812]; benchmarks 6; direct benchmarks 3
`moonshot/kimi-k2-6`	0.276595	[-0.076158, 0.629349]	0.00390625	Unavailable	software_engineering: -0.680378733; interval [-1.646439083, 0.285681617]; benchmarks 4; direct benchmarks 1
`deepseek/deepseek-v4-pro`	0.06661	[-0.195962, 0.329182]	0.0	0.06864	software_engineering: -1.255451879; interval [-1.974539251, -0.536364507]; benchmarks 5; direct benchmarks 2
`google/gemini-3-1-pro-preview`	0.0	[-0.255176, 0.255176]	0.0	0.128	software_engineering: -1.437870949; interval [-2.136704195, -0.739037704]; benchmarks 7; direct benchmarks 2

Thin evidence: undetermined

Model	Score	Score interval	P(best)	Cost per task, USD	Capability estimates and intervals
`openai/gpt-6-sol`	0.628205	[-0.099663, 1.356073]	0.1328125	0.12	software_engineering: 0.282550248; interval [-1.710810354, 2.275910849]; benchmarks 1; direct benchmarks 0
`qwen/qwen3-8-max-0902`	0.626192	[-0.101676, 1.35406]	0.16015625	0.104	software_engineering: 0.277037196; interval [-1.716323405, 2.270397798]; benchmarks 1; direct benchmarks 0
`qwen/qwen3-8-flash-next`	0.565178	[-0.16269, 1.293046]	0.10546875	Unavailable	software_engineering: 0.109941805; interval [-1.883418797, 2.103302407]; benchmarks 1; direct benchmarks 0
`deepseek/deepseek-v3-1`	0.500692	[-0.118844, 1.120229]	0.0546875	Unavailable	software_engineering: -0.066660632; interval [-1.763340886, 1.630019622]; benchmarks 2; direct benchmarks 0
`meta/muse-spark-1-1`	0.424279	[-0.193596, 1.042154]	0.046875	0.067	software_engineering: -0.275927718; interval [-1.968057691, 1.416202254]; benchmarks 3; direct benchmarks 1
`google/gemini-2-5-flash`	0.288934	[-0.258471, 0.836339]	0.01953125	0.022	software_engineering: -0.646587938; interval [-2.145727294, 0.852551419]; benchmarks 3; direct benchmarks 0
`microsoft/phi-4`	0.033169	[-0.586368, 0.652705]	0.00390625	Unavailable	software_engineering: -1.347034704; interval [-3.043714958, 0.34964555]; benchmarks 2; direct benchmarks 0

Unknown requirements

Some candidates may qualify. Missing facts:

software_engineering

Engine warnings

not_separable

Omitted result rows: models 18; offerings 22. The bands above include all ranked models.

Decision dec_0180e89bde9bae9adb9de365 on public snapshot snap_05a963d2477ac2fb.

What this template teaches

With no cost term, the ranking is capability alone; the canvas still shows what each candidate costs.

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