As checked 2026-10-25: A model supporting a long context window does not guarantee uniform token rates at every prompt length. This is a dated economics guide, not a permanent price guarantee or a personal hands-on benchmark.
Scope: global English reference; USD list prices where shown; local eligibility, currency, taxes, negotiated terms and model release dates require separate confirmation. Dynamic source-page prices have a check date, not a verified initial effective date.
Context capacity is not a flat-price entitlement
A model supporting a long context window does not guarantee uniform token rates at every prompt length. Google’s Gemini 3.1 Pro Preview documentation distinguishes rates for prompts at or below 200,000 tokens versus prompts above that threshold. OpenAI’s GPT-5.4 and GPT-5.5 model pages say prompts above 272,000 input tokens trigger higher rates for the session. The threshold wording matters: do not apply only the higher rate to the tokens beyond the boundary when a provider explicitly says the entire session is repriced.
Example of a threshold effect
Imagine 199,000 input tokens and 5,000 output tokens charged under a short-context price schedule. Adding 2,000 more source tokens may push the request into a higher rate category that applies to every input and output token, not just the extra 2,000. A spreadsheet that multiplies the first 200,000 by one rate and the remaining 1,000 by another will be wrong if the vendor uses whole-request thresholds. The correct selector reads the vendor’s current definition, and is tested against a response whose usage fields can be audited.
Reduce unnecessary context before selecting a price tier
Use retrieval to choose relevant passages rather than concatenating every article, prune repeated instructions, and summarize stable historical context without hiding uncertainty. Chunking can cost more if it repeats the same instructions or increases verification calls, so compare total token consumption and accuracy. Caching is not a universal escape: some providers count cached tokens in threshold decisions even when the cache hit is discounted.
Model-specific and regional qualifications
Do not extrapolate GPT-5.5’s threshold to all OpenAI models, or Gemini 3.1 Pro’s 200k split to every Gemini Flash variant. Preview identifiers can differ from stable releases, and cloud resellers can price differently. Document the full model ID, request size metric, region, tier, input and output rate schedules, and whether the threshold changes both sides of the bill.
What to publish and what to leave unknown
If the rule or base rate is not verifiably documented, show the method without inventing a numeric threshold. A long-context purchase decision should weigh retrieval quality, response completeness, latency and review time as well as raw token price. Record a representative distribution of prompt sizes; a median smaller than the threshold does not rule out expensive rare requests.
Investigate high-percentile prompts
Report p50, p90, p99 and maximum prompt length by request type, not just an overall monthly mean. A knowledge-base agent can have a median of 20,000 tokens while a handful of massive retrievals drive cost exposure. Add a rate selector that records the measured prompt length and whether any context threshold was crossed; preserve the selected price ID. If the provider’s policy counts cached prefix tokens toward the threshold, a repeated long prompt may still occupy an expensive bracket despite discounted reads. Never interpolate an undocumented boundary.
Architecture alternatives require QA
Chunk-and-summarize designs can reduce a single oversized prompt, but may lose key evidence or require several additional outputs. A retrieval-first system may be cheaper yet miss critical details if the index is poor. Define a set of questions with source-backed expected answers, evaluate the full long-context prompt against a retrieval design, and compare accepted-answer costs rather than raw token rates. Keep human review for financial, legal or historical facts. This evaluation should be run with the exact model versions being considered; do not import vendor claims about context quality as independent results.
Decision checklist
- Identify the specific plan or API endpoint, model version, region and currency before pricing a job.
- Measure input, output, caching, generated media and failed-job units separately.
- Check the official pricing page again before procurement, especially if the SKU has no effective-date notice.
- Keep source evidence and a timestamped price ID; do not treat unknowns as zero.
Pricing evidence boundary
This cost discussion concerns Multi-vendor. Reproduce a quote only with its exact model or plan identifier, chargeable unit, billing region, processing mode and timestamp; a changing vendor rate card is not a permanent price guarantee. If any one of these dimensions is missing, keep the associated rate and worked scenario under editorial review instead of filling the gap with an estimate.
Sources & Further Reading
- gpt-5.5 text-token standard list rates and threshold above 272k input tokens — official first-party source, accessed 2026-10-25; original effective date not specified in source.
- gpt-5.4 text-token standard list rates and threshold above 272k input tokens — official first-party source, accessed 2026-10-25; original effective date not specified in source.
- Gemini Developer API shows tier/model/context/modality-specific rates and cache storage — official first-party source, accessed 2026-10-25; original effective date not specified in source.
- xAI Grok API short/long context token-rate table — official first-party source, accessed 2026-10-25; original effective date not specified in source.