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.

High volume, quality first

Find a text model for many short assistant requests where response quality still leads the ranking.

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

Short requests at scale; quality leads, cost counts.

Requirements

model.class = text-generator: The workload needs generated text responses.

Preferences

chat_preference: Prefer assistants whose responses people rate more highly.
-offering.cost_per_task: At volume, cost still matters as the second term.

Applied template spec

spec_version

where

model.class = text-generator

optimize

weights

chat_preference: 0.7
-offering.cost_per_task: 0.3

task_tokens

input: 2000
output: 500

Worked answer

Decision status: partial.

Tied on the evidence

anthropic/claude-opus-5-5
google/gemini-3-8-flash
qwen/qwen3-8-max-0902
openai/gpt-6-sol

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: google/gemini-3-8-flash
Most independently measured tie-breaker: google/gemini-3-8-flash

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`	0.884932	[0.71708, 1.052783]	0.4921875	0.018	chat_preference: 1.584232494; interval [0.565758585, 2.602706403]; benchmarks 1; direct benchmarks 1
`openai/gpt-6-sol`	0.784717	[0.616865, 0.952569]	0.18359375	0.009	chat_preference: 0.602126414; interval [-0.416347496, 1.620600323]; benchmarks 1; direct benchmarks 1
`qwen/qwen3-8-max-0902`	0.79648	[0.628628, 0.964331]	0.15234375	0.007	chat_preference: 0.59037787; interval [-0.428096039, 1.60885178]; benchmarks 1; direct benchmarks 1
`google/gemini-3-8-flash`	0.80065	[0.771534, 0.829766]	0.0546875	0.003375	chat_preference: 0.465029132; interval [0.288363151, 0.641695113]; benchmarks 24; direct benchmarks 24

Rest band

Model	Score	Score interval	P(best)	Cost per task, USD	Capability estimates and intervals
`google/gemini-3-7-flash`	0.790269	[0.761153, 0.819385]	0.03515625	0.003375	chat_preference: 0.402038706; interval [0.225372725, 0.578704687]; benchmarks 24; direct benchmarks 24
`meta/muse-spark-1-3`	0.742647	[0.704627, 0.780668]	0.00390625	0.004625	chat_preference: 0.165034925; interval [-0.065661987, 0.395731837]; benchmarks 13; direct benchmarks 13
`zhipu/glm-5-3`	0.738166	[0.711666, 0.764666]	0.0	0.005	chat_preference: 0.153431285; interval [-0.007361761, 0.31422433]; benchmarks 29; direct benchmarks 29
`xai/grok-4-7`	0.732945	[0.565093, 0.900796]	0.078125	0.007	chat_preference: 0.204866691; interval [-0.813607219, 1.2233406]; benchmarks 1; direct benchmarks 1
`anthropic/claude-opus-4-6`	0.720795	[0.697, 0.74459]	0.0	0.0225	chat_preference: 0.775317472; interval [0.630936802, 0.919698141]; benchmarks 36; direct benchmarks 36
`anthropic/claude-opus-4-7`	0.703004	[0.678522, 0.727486]	0.0	0.0225	chat_preference: 0.667370368; interval [0.518820066, 0.81592067]; benchmarks 34; direct benchmarks 34
`google/gemini-3-1-pro-preview`	0.680454	[0.655972, 0.704936]	0.0	0.01	chat_preference: 0.01105026; interval [-0.137500042, 0.159600562]; benchmarks 34; direct benchmarks 34
`anthropic/claude-opus-5`	0.67912	[0.654638, 0.703602]	0.0	0.0225	chat_preference: 0.522450073; interval [0.373899771, 0.671000375]; benchmarks 34; direct benchmarks 34
`meta/muse-spark-1-1`	0.666492	[0.633792, 0.699191]	0.0	0.004625	chat_preference: -0.297052895; interval [-0.4954636, -0.098642189]; benchmarks 19; direct benchmarks 19
`openai/gpt-5-6-sol`	0.659942	[0.635085, 0.684798]	0.0	0.018	chat_preference: 0.219061677; interval [0.068242727, 0.369880626]; benchmarks 33; direct benchmarks 33
`google/gemini-3-5-flash`	0.651024	[0.626542, 0.675506]	0.0	0.0075	chat_preference: -0.271422803; interval [-0.419973105, -0.122872501]; benchmarks 34; direct benchmarks 34
`zhipu/glm-5-2`	0.644935	[0.618436, 0.671435]	0.0	0.005	chat_preference: -0.412265302; interval [-0.573058347, -0.251472257]; benchmarks 29; direct benchmarks 29
`google/gemini-2-5-flash`	0.639407	[0.613343, 0.665472]	0.0	0.00185	chat_preference: -0.576719849; interval [-0.734872721, -0.418566977]; benchmarks 34; direct benchmarks 34
`openai/gpt-5-4`	0.635099	[0.611304, 0.658894]	0.0	0.0125	chat_preference: -0.16025079; interval [-0.304631459, -0.01587012]; benchmarks 36; direct benchmarks 36
`anthropic/claude-fable-5`	0.609594	[0.585112, 0.634076]	0.0	0.045	chat_preference: 1.035674964; interval [0.887124662, 1.184225266]; benchmarks 34; direct benchmarks 34
`deepseek/deepseek-v4-pro`	0.584686	[0.558187, 0.611186]	0.0	0.00462	chat_preference: -0.793630284; interval [-0.954423329, -0.632837239]; benchmarks 29; direct benchmarks 29
`anthropic/claude-fable-5-1`	0.566597	[0.540962, 0.592231]	0.0	0.045	chat_preference: 0.774780684; interval [0.619237933, 0.930323435]; benchmarks 31; direct benchmarks 31
`openai/gpt-6-astra`	0.445507	[0.417527, 0.473486]	0.0	0.045	chat_preference: 0.040043976; interval [-0.12972742, 0.209815373]; benchmarks 26; direct benchmarks 26

Unknown requirements

Some candidates may qualify. Missing facts:

chat_preference
offering.cost_per_task

Engine warnings

not_separable

Omitted result rows: models 11; offerings 18. The bands above include all ranked models.

Decision dec_f0ebf8d327afab69afc0e656 on public snapshot snap_05a963d2477ac2fb.

What this template teaches

Short task tokens make the cost term reflect one small request, not one long agent run.

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