UK-based independent AI efficiency projectEvidence before claims · 2026

Token Optimisation

AI efficiency · measured
Discuss a diagnostic

AI Context Cost Calculator

Official API prices · measured subscription usage · local calculation

Compare API model rates or subscription workload

Compare current and target request contexts using measured token counts or explicit assumptions. Or use subscription mode and enter measured jobs, machine time and usage. Calculations run in your browser; no AI, tokenizer, pricing or FX API is called.

Your AI service / model

API models show official input, cache-read and output prices. Subscription products use measured workload mode, not per-token prices.
Native provider rates remain visible; converted totals show source and date.

Subscription / measured usage mode

No official per-million-token rate is inferred for a ChatGPT subscription. Derived figures use the cost and measurements shown below.

OBSERVED · USER-RECORDED ChatGPT Pro 100 cost: £89, purchase date 2026-10-02; benchmark IN PROGRESS. Billing period is unverified.

Official plans list Pro from £89/month; this run uses the user-recorded £89 charge, not a per-token price.

Available workload evidence (scoped snapshots)

Local Bridge/OpenCode read-only capture, 2026-10-02 13:01 UTC: zero local usage rows, provider attempts, Bridge auth events and verified production outputs since the stated Pro 100 start; one queued job, zero running, queue paused. This ledger does not cover other clients; total account usage and maximum context are unknown.

Separate OpenCode auth diagnostic: after the owner-reported plan change, redacted app logs showed invalidation events across two sessions. The measured interval from first logged failure to auth-store refresh was 6.39 minutes; no shared backend restart was observed. Later provider streams appeared in one session and none in the other; a verified job outcome was not established. Human reauthentication minutes are unknown.

These snapshots have different scopes. They are not summed into one total. Synthetic WAIT_AUTH tests are not production usage and are not included in the calculator.

Related reading: heartbeat versus model progress.

Observed £89 purchase on 2026-10-02; period not verified.
Accepted/verified jobs only.
Same cost window; idle time excluded.
Provider/session telemetry.
Breakdown only; never added twice.
Same benchmark window.

Historical Plus plan usage and invoices were not found in reviewed local evidence. The £20 current Plus list price is not treated as a historical baseline.

Results

Method and boundaries

API spend = uncached input × input price + cached input × cache-read price + output × output price, divided by 1M. Subscription cost/job, cost/useful hour and effective cost/1M measured tokens appear only when those measurements exist.

USD conversion uses ECB reference rates dated 1 October 2026 (1 EUR = USD 1.1298; GBP 0.85373), not a live quote. View ECB source.

Related: Tools directory · AI Costs · Methodology · Experiments · Diagnostic enquiry.