AI model platform / transparency / Terminal Green + Swiss Modern
Choose a model with the card, cost, evals, and status in view.
Model Ledger turns model selection into an inspectable system: versioned model cards, benchmark methods, API examples, pricing units, safety limits, and incident history.
Text, image input, structured output
POST /v1/responses
{ "model": "ledger-large-2026-05" }
8models
6eval suites
99.94%API
Model card
AI platform pages need versioned evidence, not only a launch claim.
ledger-large-2026-05. Text, image input, structured output. 128k context, 16k max output.
Intended use
Research synthesis, code review, support automation, long-document analysis
Limitations
Not for medical diagnosis, legal advice, autonomous high-stakes decisions, or unsupervised financial recommendations.
Changelog
May 2026: better citation grounding, lower tool-call latency, deprecated ledger-large-2026-02.
Benchmarks with method
Scores need datasets, baselines, dates, and caveats in the same view.
Benchmarks here show method notes beside the score so the table cannot become marketing confetti.
Reasoning
83.4Internal 2,400-task suite, May 2026, compared with ledger-large-2026-02 baseline 79.1
Code repair
71.8Public issue set, pass@1, contamination screened, baseline 68.0
Long context
92.2Needle and multi-source synthesis suite, 128k context
Pricing and calculator
Developers need billing units before they can choose the model.
Example: 800k input, 120k output, and 300k cached input estimates to $3.23 before batch discounts.
Input
$2.40 / 1M tokensShown with rate-limit and enterprise notes on the pricing page.
Output
$9.60 / 1M tokensShown with rate-limit and enterprise notes on the pricing page.
Cached input
$0.60 / 1M tokensShown with rate-limit and enterprise notes on the pricing page.
Batch
50% discount, 24h windowShown with rate-limit and enterprise notes on the pricing page.
Status, safety, docs
Transparency needs component status, safety boundaries, docs, and evals to stay synchronized.
A green global dot is not enough for developer operations or procurement.
API inference
Operational99.94% last 30 days
Console
DegradedUpload queue delay, opened 09:12 UTC
Embeddings
OperationalNo active incidents
External red-team summaryHigh-impact decision boundaryData retention controlsVulnerability disclosure
Decision path
Compare models
Open model card before you commit.
An AI developer-platform transparency sample with model cards, benchmark methodology, pricing calculator, safety notes, evals, changelog, docs, and component-level status.
AI model transparency / developer platform site