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

Find a text model that can accept million-token documents and produce strong written analysis.

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

Million-token context; quality first, cost as a tie-breaker.

Requirements

model.class = text-generator: The result must be generated as text.
model.context_window >= 1000000: The model must accept a million-token document in one request.

Preferences

writing: Prefer clear, high-quality long-form writing.
reasoning: Prefer stronger multi-step analysis of the document.
-offering.cost_per_task: Use cost as a smaller tie-breaker after writing and reasoning quality.

Applied template spec

spec_version

where

model.class = text-generator
model.context_window >= 1000000

optimize

weights

writing: 0.5
reasoning: 0.3
-offering.cost_per_task: 0.2

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
`google/gemini-3-7-flash`	0.791458	[-0.257187, 1.840103]	0.1015625	0.045	reasoning: 0.463255629; interval [-1.246777358, 2.173288616]; benchmarks 2; direct benchmarks 0 writing: 0.434597365; interval [-1.064541992, 1.933736721]; benchmarks 3; direct benchmarks 0
`google/gemini-3-8-flash`	0.78012	[-0.216673, 1.776913]	0.08203125	0.045	reasoning: 0.570955383; interval [-0.928430754, 2.070341519]; benchmarks 3; direct benchmarks 0 writing: 0.344204382; interval [-1.154934975, 1.843343738]; benchmarks 3; direct benchmarks 0
`anthropic/claude-opus-4-6`	0.741334	[-0.257456, 1.740125]	0.06640625	0.3	reasoning: 0.197361213; interval [-1.310140483, 1.704862909]; benchmarks 3; direct benchmarks 0 writing: 0.690665563; interval [-0.808473794, 2.189804919]; benchmarks 3; direct benchmarks 0
`anthropic/claude-fable-5-1`	0.67071	[-0.373527, 1.714948]	0.04296875	0.6	reasoning: 0.710115622; interval [-0.98201435, 2.402245595]; benchmarks 2; direct benchmarks 0 writing: 0.478747567; interval [-1.02039179, 1.977886923]; benchmarks 3; direct benchmarks 0
`anthropic/claude-fable-5`	0.634924	[-0.414579, 1.684427]	0.0625	0.6	reasoning: 0.039498036; interval [-1.674020333, 1.753016406]; benchmarks 2; direct benchmarks 0 writing: 0.787525395; interval [-0.711613961, 2.286664752]; benchmarks 3; direct benchmarks 0
`google/gemini-3-1-pro-preview`	0.623031	[-0.37576, 1.621821]	0.0625	0.128	reasoning: 0.22241102; interval [-1.285090676, 1.729912715]; benchmarks 3; direct benchmarks 0 writing: 0.245370198; interval [-1.253769159, 1.744509554]; benchmarks 3; direct benchmarks 0
`openai/gpt-5-6-sol`	0.580157	[-0.472287, 1.632601]	0.05859375	0.24	reasoning: 0.148460441; interval [-1.577008032, 1.873928913]; benchmarks 2; direct benchmarks 0 writing: 0.282841067; interval [-1.216298289, 1.781980424]; benchmarks 3; direct benchmarks 0
`anthropic/claude-opus-4-7`	0.569051	[-0.42974, 1.567841]	0.05859375	0.3	reasoning: -0.117498278; interval [-1.624999974, 1.390003417]; benchmarks 3; direct benchmarks 0 writing: 0.464308039; interval [-1.034831318, 1.963447396]; benchmarks 3; direct benchmarks 0
`anthropic/claude-opus-5`	0.566419	[-0.484461, 1.617299]	0.04296875	0.3	reasoning: 0.45386262; interval [-1.265252052, 2.172977292]; benchmarks 2; direct benchmarks 0 writing: 0.122128685; interval [-1.377010672, 1.621268041]; benchmarks 3; direct benchmarks 0
`meta/muse-spark-1-3`	0.53194	[-0.798338, 1.862218]	0.109375	0.067	reasoning: 0.211605235; interval [-1.789205691, 2.212416162]; benchmarks 1; direct benchmarks 0 writing: -0.018325411; interval [-2.019136338, 1.982485515]; benchmarks 1; direct benchmarks 0
`google/gemini-3-5-flash`	0.518371	[-0.538896, 1.575638]	0.078125	0.096	reasoning: 0.140878753; interval [-1.604182337, 1.885939842]; benchmarks 2; direct benchmarks 0 writing: 0.015804984; interval [-1.483334372, 1.514944341]; benchmarks 3; direct benchmarks 0
`zhipu/glm-5-3`	0.46333	[-0.584231, 1.510891]	0.0546875	0.0736	reasoning: -0.105195063; interval [-1.810826758, 1.600436632]; benchmarks 2; direct benchmarks 0 writing: 0.009738247; interval [-1.489401109, 1.508877604]; benchmarks 3; direct benchmarks 0
`openai/gpt-6-astra`	0.454539	[-0.542479, 1.451557]	0.04296875	0.6	reasoning: 0.323374241; interval [-1.176927713, 1.823676194]; benchmarks 3; direct benchmarks 0 writing: 0.189832533; interval [-1.309306823, 1.68897189]; benchmarks 3; direct benchmarks 0
`openai/gpt-5-4`	0.408376	[-0.60046, 1.417212]	0.0234375	0.16	reasoning: 0.230107427; interval [-1.318205455, 1.77842031]; benchmarks 3; direct benchmarks 0 writing: -0.244266234; interval [-1.743405591, 1.254873122]; benchmarks 3; direct benchmarks 0
`meta/muse-spark-1-1`	0.329212	[-1.001067, 1.65949]	0.05859375	0.067	reasoning: -0.343673105; interval [-2.344484032, 1.657137821]; benchmarks 1; direct benchmarks 0 writing: -0.176054846; interval [-2.176865772, 1.824756081]; benchmarks 1; direct benchmarks 0
`zhipu/glm-5-2`	0.306316	[-0.747665, 1.360297]	0.03515625	0.0736	reasoning: -0.275228659; interval [-2.006940496, 1.456483179]; benchmarks 2; direct benchmarks 0 writing: -0.265292672; interval [-1.764432028, 1.233846685]; benchmarks 3; direct benchmarks 0
`deepseek/deepseek-v4-pro`	0.191481	[-0.863308, 1.246271]	0.01953125	0.06864	reasoning: -0.508628162; interval [-2.24362481, 1.226368486]; benchmarks 2; direct benchmarks 0 writing: -0.406605901; interval [-1.905745257, 1.092533456]; benchmarks 3; direct benchmarks 0

Unknown requirements

Some candidates may qualify. Missing facts:

model.context_window
offering.cost_per_task
reasoning
writing

Engine warnings

not_separable
proxy_evidence_only

Omitted result rows: models 7; offerings 10. The bands above include all ranked models.

Decision dec_da950856dd6ff0655d97ff6a on public snapshot snap_05a963d2477ac2fb.

What this template teaches

Context capacity is a Must; writing, reasoning, and cost remain trade-offs expressed as Prefers.

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