Pricing methodology & sources

1. How prices get on this site

• Manual verification. A human opens the provider's official pricing page, copies the figures, and stamps the date you see above. There is no live price feed — we deliberately don't claim realtime sync.
• Weekly re-checks. An automated job re-opens the official pages every week and flags any drift; the stamp above moves only after a human confirms the new numbers.
• One data layer. Every calculator on this site reads from a single pricing file — a price can't be $2.00 on one page and $2.20 on another.
• No invented numbers. If a provider hasn't published a rate (or we haven't verified it), it doesn't appear on this site. Period.

2. Current model architecture and previous models

The homepage lineup is GPT-6 Luna, GPT-6.1 Sol and GPT-6 Astra. Each model records its API id, family, variant, status, source, verified date, pricing bands and processing availability. GPT-6 Sol is a previous model replaced by GPT-6.1 Sol; its SEO page and canonical remain available.

GPT-6.1 Sol calculator · Current vs previous Sol

Short vs long context

For each configured GPT-6 model, input above 272,000 tokens uses the long-context rates for the entire request. Exactly 272,000 tokens is short context. Fresh input, cache reads and writes double; output uses 1.5× the short-context rate. Thresholds are model-specific data, not a global assumption.

3. Processing modes and regional processing

Standard is 1×, Fast 2×, Flex 0.5× and Batch 0.5× for the verified OpenAI models. GPT-6 Astra and GPT-6.1 Sol also offer Ultrafast mode at 6× Standard — for Astra, OpenAI publishes it as its own price table ($60/$6/$75/$300 per 1M short-context tokens, $120/$12/$150/$450 long-context); for 6.1 Sol the model page publishes the 6× multiplier directly ($12/$0.60/$15/$60 short-context, $24/$1.20/$30/$90 long-context). Every figure is exactly six times the Standard rate in both context bands, so the data layer models Ultrafast as a 6× processing mode rather than a separate price list. Availability comes from each model's configuration. EU regional processing adds 10% where eligible. EU plus Fast is available on Luna, Sol and 6.1 Sol, but blocked on Astra; EU plus Ultrafast is available on 6.1 Sol only (Astra Ultrafast supports US data residency and global processing only). The UI explicitly switches an incompatible mode to Standard and explains the change. Account eligibility and tool restrictions still apply.

4. Calculation engine and input buckets

Every calculator uses the unified pricing-data.js engine. Fresh input share = 100 − cached-read share − cache-write share. Cached reads and cache writes replace fresh-input tokens; they are never added to the same tokens twice. The engine rejects shares over 100%; UI controls cap writes to the remaining share and explain the adjustment.

perRequest = (freshTokens × inputRate
            + cachedReadTokens × cachedReadRate
            + cacheWriteTokens × cacheWriteRate
            + outputTokens × outputRate) / 1,000,000
             × processingMultiplier × regionalMultiplier
perMonth = perRequest × requestsPerMonth
perDay = perMonth / 30
perYear = perMonth × 12

Cost components returned by the engine include processing and regional multipliers. Comparisons use unrounded costs with the same workload; unavailable combinations are labeled rather than priced. Zero selected cost has no relative comparison. Published rates exclude tools, taxes, storage, negotiated pricing and eligibility rules. Output tokens should include billed reasoning tokens.

5. Exactly what the calculator computes

For each request, split total input tokens into fresh input, cached reads and cache writes. Select the model's short- or long-context rates using total input tokens before splitting the buckets; a configured long-context band applies to the entire request when its threshold is exceeded.

Three worked examples, using the verified rates in the tables above:

• Cache split (GPT-6.1 Sol, short context). 100,000 input tokens per request: 30,000 fresh at $2.00/1M = $0.06; 60,000 cached reads at $0.10/1M = $0.006; 10,000 cache writes at $2.50/1M = $0.025; plus 2,000 output tokens at $10.00/1M = $0.02. Per-request cost = $0.111; at 10,000 requests/month = $1,110/month.
• Long context (Luna). 300,000 input tokens exceeds Luna's 272,000-token threshold, so the entire request bills at long-context rates — not just the tokens above the threshold: 300,000 × $0.20/1M = $0.06 input, plus 5,000 output tokens × $0.75/1M = $0.00375. Per-request ≈ $0.0638.
• Stacked multipliers (GPT-6.1 Sol, EU + Flex). Multipliers apply once to every cost component, including output. A $0.10 base per-request cost with Flex (0.5×) and EU regional processing (1.1×) becomes $0.10 × 0.5 × 1.1 = $0.055. EU + Fast is also valid on this model (2× then 1.1×); on Astra, where EU + Fast is blocked, the calculator falls back to Standard and says so.

Key point: cached reads and cache writes replace fresh input for their respective portions; the three input buckets never overlap. For configured OpenAI models, the processing multiplier is 1× Standard, 2× Fast, 0.5× Flex/Batch, or 6× Ultrafast where that mode is published (Astra and 6.1 Sol). The regional multiplier is 1.1× for eligible EU processing, otherwise 1×. Each multiplier applies once to every cost component, including output, and is already included in per_request.

6. Other providers: Claude & Gemini

Our comparison pages also show verified Anthropic and Google prices — with one honest caveat: caching works differently per provider, so cross-provider comparisons are close estimates, not exact invoices.

• Anthropic: the "cached input" figure is the cache-hit rate — 0.1× base on most models, 0.05× on Opus 5.5 ($0.20) and Sonnet 5.5 ($0.10), 0.025× on Fable 5.1 / Mythos 5.1 ($0.25). Cache writes cost 1.25–2× the base input price and are not modeled in our calculators — heavy cache churn will cost more than shown. Haiku 5.5 (marked † above) is the one Anthropic model priced by prompt length: prompts over 100,000 tokens bill the whole request at $0.50 input / $0.05 cache hit / $2.50 output per 1M, and the calculator applies that band automatically.
• Google: figures are short-context rates (prompt ≤ 200K tokens). Gemini 2.5 Pro charges more for longer prompts; Flash / Flash-Lite text rates shown (audio input costs more).
• Cross-provider "cached %" comparisons assume the same reuse pattern — in practice each provider's caching rules differ. Treat these as directional.

7. What we won't do

• No "live" or "realtime" price claims — the stamp at the top is the truth.
• No invented rates for tiers, regions, or models we haven't verified.
• No affiliation claims — this is an independent tool, not affiliated with or endorsed by OpenAI, Anthropic, or Google.
• No quality judgments dressed as math — comparisons here are strictly cost arithmetic.

8. Changelog

• October 10, 2026 — Added Claude Haiku 5.5 (launched October 7) to the data layer and the comparison widgets: $0.10 input / $0.01 cache hit / $0.50 output per 1M for prompts up to 100,000 tokens, stepping to $0.50 / $0.05 / $2.50 when a prompt exceeds 100,000 tokens (whole-request prompt-length pricing). All figures verified line-by-line against Anthropic's official pricing page, including the long-context and prompt-caching sections; cache-write surcharges remain unmodeled per our standing policy. The October 8 blog comparison previously noted Haiku 5.5 was not yet a calculator model; it now is, and the article has been updated to say so.
• October 10, 2026 — Added Ultrafast mode (6× Standard) for GPT-6.1 Sol: OpenAI's model page now publishes the tier (short-context $12.00/$0.60/$15.00/$60.00 per 1M by the 6× multiplier), superseding the October 3 correction that removed it as unpublished. Also corrected EU regional availability per the current model pages and Fast/Ultrafast guides: Fast now combines with EU data residency on Luna, Sol and 6.1 Sol, and 6.1 Sol Ultrafast supports EU residency; Astra Fast/Ultrafast remain blocked with EU. All per-token rates for OpenAI (Luna / 6.1 Sol / Astra short+long), Anthropic (Opus 5.5 / Sonnet 5.5 / Sonnet 5 / Haiku 4.5 / Fable 5.1 / Mythos 5.1) and Google (Gemini 2.5 Pro / Flash / Flash-Lite) re-verified unchanged against official pricing pages the same day.
• October 8, 2026 — Claude Sonnet 5.5 cache-hit rate updated from $0.20 to $0.10 per 1M (now 0.05× base, per Anthropic's official pricing table and prompt-caching notes); the vs-Claude comparison example and FAQ updated to match. All other OpenAI, Anthropic and Google rates re-verified unchanged the same day.
• October 3, 2026 — Added Ultrafast mode (6× Standard) for GPT-6 Astra to the data layer and calculators. Rates copied from OpenAI's official pricing page (Astra $60/$6/$75/$300 short-context, $120/$12/$150/$450 long-context per 1M — the published Ultrafast table lists Astra only) and confirmed against the official Ultrafast guide; Ultrafast is blocked with EU data residency. All existing OpenAI, Anthropic and Google rates re-verified unchanged the same day. Correction (same day): an earlier edit the same day briefly listed Ultrafast for GPT-6.1 Sol using rates derived from its Standard prices; OpenAI does not publish an Ultrafast tier for 6.1 Sol, so those entries were removed.
• September 30, 2026 — Added Claude Sonnet 5.5 ($2.00/$0.20/$10.00), Fable 5.1 and Mythos 5.1 ($10.00/$0.25/$50.00) to the data layer, the vs-Claude comparison page, and the table above. Rates verified against Anthropic's official pricing page, including the special 0.025× cache-hit rate on Fable/Mythos 5.1. GPT-6.1 Sol long-context rates ($4.00/$0.20/$15.00) also verified.
• September 26, 2026 — Initial publication. GPT-6 short/long-context rates, tier multipliers, Claude (Opus 5.5 / Sonnet 5 / Haiku 4.5) and Gemini (2.5 Pro / Flash / Flash-Lite) rates verified against official pricing pages.