API cost, timing and quality.
See which actions use a model, how the report budgets each journey, and what the timing records establish.
Cost tables use the report’s price snapshot from 11 October 2026 and explicit token budgets. Actual charges follow billed usage and the service provider’s rates.
What would this journey cost?
Change how often each action happens. A photo action includes both observation and matching; one guidance sentence uses one text call.
Use whole numbers from 0 to 100 for actions and 1 to 1,000,000 for journeys.
USD / journey
Total USD · 1,000 journeys
Model calls / journey
7,050 input tokens + 1,090 output tokens per journey.
Model inference budget. Maps, hosting, storage and browser speech have separate service-cost categories. Page viewing, route search, confirmation and session synchronization add zero GenAI calls.
Prices turn tokens into charges.
A token is a unit of model input or generated output. Input includes the instructions, context and image tokens. Output includes billable reasoning tokens when the selected mode produces them.
| Model | Input | Output |
|---|---|---|
| Qwen3.8-Max | $2.00 | $6.00 |
| Qwen3-VL-Plus | $0.20 | $1.60 |
Historical rates reproduced from report Table 4; vision input tier ≤ 32,000 tokens. T = Qwen3.8-Max, V = Qwen3-VL-Plus throughout the call table.
Charge (USD) = (input tokens × input price + output tokens × output price) / 1,000,000
Each action has a call budget.
Reading a photo and matching its evidence use different model calls. Local map retrieval and deterministic ranking sit between those calls.
| GenAI call | Model | Input tokens | Output tokens | Total tokens | USD / call |
|---|---|---|---|---|---|
| Preference parsing | T | 1,200 | 180 | 1,380 | $0.003480 |
| Landmark observation | V | 2,400 | 320 | 2,720 | $0.000992 |
| Obstacle observation | V | 2,600 | 380 | 2,980 | $0.001128 |
| Landmark evidence matching | T | 1,800 | 240 | 2,040 | $0.005040 |
| Obstacle evidence matching | T | 2,200 | 260 | 2,460 | $0.005960 |
| Cantonese guidance | T | 450 | 70 | 520 | $0.001320 |
| User action | Model sequence | Calls | USD / action |
|---|---|---|---|
| Understand preferences | Preference parsing | 1 | $0.003480 |
| Locate from a landmark photo | Landmark observation → Landmark evidence matching | 2 | $0.006032 |
| Match an obstacle photo | Obstacle observation → Obstacle evidence matching | 2 | $0.007088 |
| Write one guidance sentence | Cantonese guidance | 1 | $0.001320 |
A journey combines the actions.
The report lists three workloads. Guidance counts include the sentences generated after route updates.
| Workload | Calls | Input tokens | Output tokens | USD / journey | USD / 1,000 journeys |
|---|---|---|---|---|---|
| Ordinary journey1 preference request · 13 guidance sentences · 0 landmark photo · 0 obstacle photo | 14 | 7,050 | 1,090 | $0.020640 | $20.640 |
| With location assistance1 preference request · 13 guidance sentences · 1 landmark photo · 0 obstacle photo | 16 | 11,250 | 1,650 | $0.026672 | $26.672 |
| With both photo workflows1 preference request · 15 guidance sentences · 1 landmark photo · 1 obstacle photo | 20 | 16,950 | 2,430 | $0.036400 | $36.400 |
Response time needs a defined scope.
The records distinguish one observed response, a configured deadline, and the workflow statistics that still need measurement. p95 means 95% of measured requests finish within that time.
| Item | Time / statistic | Evidence type | What the record covers |
|---|---|---|---|
| Text API connectivity sample | About 3.4 s | Observed sample | One successful solar-energy text request, HTTP 200; live integration note dated 2026-10-04. This request differs from the navigation workflows. |
| Model request timeout | 60 s / call | Configuration | The provider aborts an individual model request after 60,000 ms. This limit describes a request deadline. |
| Navigation workflow latency | Median / p95 pending | Measurement pending | The saved integration note records successful workflow outcomes without a latency distribution. Photo workflows contain two sequential model calls plus local processing. |
Text workflow1 model call + local validation
Photo workflowVision → local retrieval → text matching → local ranking
The 3.4-second sample provides connectivity evidence. It does not supply navigation-workflow averages. Measure each complete workflow from request submission to usable result; record sample count, median, p95 and failures together.
Quality has explicit acceptance targets.
The report defines the following thresholds for evaluation. These percentages are acceptance targets; the tables do not contain measured accuracy results.
Show component and workflow target tables
| Metric | Correctness criterion | Target |
|---|---|---|
| Preference field accuracy | Each requested field matches its reference value. | 97.5% |
| Complete-request accuracy | All requested fields and retained profile values match. | 94.0% |
| Numeric-condition accuracy | Slope, width, units, and normalized values match. | 99.0% |
| Landmark-name accuracy | The photo observation identifies the annotated name. | 94.0% |
| Shop-identifier accuracy | The normalized shop identifier matches exactly. | 95.0% |
| Visible-floor accuracy | The observation matches the annotated visible floor. | 96.0% |
| Obstacle-presence accuracy | The predicted presence or absence matches the image label. | 95.0% |
| Obstacle-type accuracy | The predicted category matches the reference category. | 93.0% |
| Location top-1 accuracy | The first candidate matches the reference public anchor. | 92.0% |
| Location top-3 accuracy | The first three candidates include the reference anchor. | 98.0% |
| Obstacle-target accuracy | The first candidate identifies the correct facility or passage. | 95.0% |
| Structured-output accuracy | The response satisfies the task schema on the first call. | 99.5% |
| Guidance fact accuracy | Action, facility, distance, and floors match the route facts. | 99.0% |
| Cantonese expression accuracy | A fluent speaker accepts grammar, wording, and action clarity. | 96.0% |
| Metric | Correctness criterion | Target |
|---|---|---|
| Language-to-route accuracy | Confirmed fields match the reference and the route satisfies them. | 92.0% |
| Photo-to-position accuracy | Confirmation selects the reference scene, floor, and public node. | 90.0% |
| Obstacle-to-reroute accuracy | Confirmation closes the correct object and the route avoids it. | 91.0% |
| Lift-required journey accuracy | Every floor change uses an eligible lift throughout the journey. | 99.0% |
| Sheltered-route accuracy | The route meets the active shelter objective and access conditions. | 98.0% |
| Guidance-to-step accuracy | Every instruction matches its active segment after progress or rerouting. | 98.0% |
Trace every number to its source.
Costs and targets follow section 23 of the technical report. Timing comes from the dated live-integration note and the provider’s request configuration.
Technical report · section 23 ↗Data snapshot · JSON ↓Budget tables · CSV ↓Report extract & timing provenance ↓
- Live API integration record · 4 October 2026 ↗
- Provider request timeout · source snapshot ↗
- Price sources cited by the report: Qwen3.8-Max ↗Qwen3-VL-Plus ↗