01
Data Sourcing
Pricing data is aggregated from live marketplace endpoints (including OpenRouter) and hand-curated from official provider API documentation and published pricing pages. We do not interpolate, estimate, or infer prices. If a provider does not publish pricing, the model is listed without a price rather than with a placeholder. Significant price movements are held for human review before publication.
02
How the Data Pipeline Runs
The schedule is fixed and automated. At 02:00 UTC a job refreshes marketplace pricing into our database and records any field that moved. At 03:00 the site rebuilds, so every static model page carries the new numbers rather than drifting from them. At 04:00 the news archive refreshes, and at 05:00 changed URLs are pushed to the search engines that accept them. The sitemap regenerates hourly. Every price movement detected since 3 May 2026 is kept as a dated record, and the model page for any given SKU shows the figure we currently hold along with the date it was last checked.
03
Blended Cost Headline
The headline blended cost shown across the Index is a default reference rate calculated as (input price × 0.7) + (output price × 0.3). This weighting reflects a generic short-prompt, longer-completion workload and is intended as a quick comparison metric, not a universal benchmark. Output-heavy workloads, agentic loops, and long-context jobs can see materially higher real costs. We always display input and output prices separately so operators can compute their own blended rates against their own traffic mix.
04
Beyond the Single Number (Roadmap)
Modern provider pricing is no longer a single input/output pair. Tiered SKUs, batch and priority modes, prompt caching, context-grounded calls, tool use, vision and audio modalities, and storage all carry distinct prices. Our paid product, in development, will move from one blended number to a SKU-aware workload modeller: pick a workload (RAG, agent, batch, voice) and Tokenando will price it across providers using the official SKU-level rates.
05
Provider Performance Layer (Roadmap)
Price alone is no longer enough. The same model can be cheaper, faster, or more reliable through different endpoints. We are building a second data layer covering provider-level latency, throughput, routing, and reliability, so the Index answers not just "what does model X cost?" but "which provider-endpoint is cheapest, fastest, or most reliable for workload Y right now?"
06
Update Frequency and History
Price data is reviewed and updated on a rolling basis as providers publish changes. The news feed aggregates from 30+ primary sources including provider blogs, industry publications, and official changelogs, refreshed every ten minutes. We are progressively building a historical pricing archive so subscribers can track per-token price movements over time. Public pages currently show the live snapshot.
07
Benchmark Scores
We do not run benchmarks. Every score on this site is the figure its publisher reported, recorded with the publisher, the date, and a link to the source. Scores are only ever compared within the same suite and the same variant, because a variant is a different benchmark: SWE-bench Pro is harder than SWE-bench Verified, so placing the two in one column would rank a stronger model below a weaker one. Suites where too few models publish a comparable figure are shown on the model page for reference and are deliberately excluded from every leaderboard and ranking.
08
Model Categorisation
Models are categorised into five tiers: Frontier (state-of-the-art, highest capability), Reasoning (chain-of-thought and extended thinking), Efficient (optimised for cost and speed), Multimodal (vision and audio capabilities), and Embedding (vector generation). Categorisation is based on provider designation and corroborated by published benchmark data.
09
Coverage Scope
We track commercially available API models across frontier labs and inference providers. Open-source weights-only models without a hosted API pricing structure are excluded. Models in private beta or with non-public pricing are listed without prices. We prioritise completeness over recency, so a model stays in the index until the provider officially deprecates the endpoint.
010
Corrections and Accuracy
Provider pricing pages, developer documentation, and launch announcements sometimes disagree with each other. Where they do, we follow the developer pricing documentation rather than the announcement prose, because the docs table is the maintained billing reference while announcement copy is written before launch and often never updated. Where a figure cannot be sourced, we publish nothing in its place. If you find a price, benchmark, or claim on this site that is wrong, email hello@tokenando.ai with the page and the source that contradicts it; we correct verified errors on the next refresh cycle.
011
Independence and Conflicts
Tokenando is independently operated. We have no commercial relationships with any of the providers tracked in our index. No provider pays for placement, ranking, or featured status. Our revenue model is subscription-based (The Compute briefing). If this changes, it will be disclosed here immediately.
