H API AI Relay Review
Review H API website uptime, official GPT and Claude pricing, API verification evidence, current offers, and clearly labeled risk notes.
API verification
AI Relay Check records model-level requests for H API and keeps failed samples. A successful request proves only that specific model and route at that time.
Pricing and billing
Official input and output rates are displayed separately. Published prices do not become verified billing costs until token usage and account balance changes are reconciled.
Decision boundary
API availability does not by itself prove model identity, long-term reliability, refund support, or balance safety. Start with a small test balance.
Independent paid test snapshot
- Tested by
- AI Relay Check review team
- Last retest
- 2026-09-29
- Protocols
- ClaudeCode / Responses
- Request success
- 77.8% (28/36)
- First byte P50 / P95
- 2.95 s / 8.40 s
- Total response P50 / P95
- 3.17 s / 18.6 s
- 7-day website reachability
- 100.0% (17)
- Sample window
- 2026-09-18 - 2026-09-29
Models tested
claude-fable-5-1 · claude-opus-5 · claude-sonnet-5 · gpt-5.5 · gpt-5.6-sol · gpt-5.6-terra · gpt-6-astra
Model-level request results
| Model | Successful / total | P50 total | Latest sample |
|---|---|---|---|
| claude-fable-5-1 | 15 / 15 | 3.14 s | 2026-09-29 |
| claude-opus-5 | 3 / 3 | 6.72 s | 2026-09-25 |
| claude-sonnet-5 | 3 / 3 | 3.65 s | 2026-09-27 |
| gpt-5.5 | 0 / 4 | - | 2026-09-23 |
| gpt-5.6-sol | 3 / 3 | 3.14 s | 2026-09-27 |
| gpt-5.6-terra | 0 / 3 | - | 2026-09-29 |
| gpt-6-astra | 4 / 5 | 3.17 s | 2026-09-25 |
A configured model without a sample is not verified. Results combine recorded requests for this model across test keys; the exact upstream group is not consistently recorded. P50 uses successful end-to-end responses only.
Provider-published model pricing
| Model | Route type | Input / 1M tokens | Output / 1M tokens | Captured |
|---|---|---|---|---|
| claude-fable-5-1 | Premium / official route | ¥16 | ¥80 | 2026-09-25 |
| claude-opus-5 | Economy / unofficial route | ¥1.5 | ¥7.5 | 2026-09-25 |
| claude-opus-5 | Premium / official route | ¥8 | ¥40 | 2026-09-25 |
| claude-sonnet-5 | Economy / unofficial route | ¥0.6 | ¥3 | 2026-09-25 |
| claude-sonnet-5 | Premium / official route | ¥3.2 | ¥16 | 2026-09-25 |
| gpt-5.5 | Economy / unofficial route | ¥0.9 | ¥5.4 | 2026-09-25 |
| gpt-5.5 | Premium / official route | ¥1.5 | ¥9 | 2026-09-25 |
| gpt-5.6-sol | Economy / unofficial route | ¥0.9 | ¥5.4 | 2026-09-25 |
| gpt-5.6-sol | Premium / official route | ¥1.5 | ¥9 | 2026-09-25 |
| gpt-5.6-terra | Economy / unofficial route | ¥0.36 | ¥2.16 | 2026-09-25 |
| gpt-5.6-terra | Premium / official route | ¥0.6 | ¥3.6 | 2026-09-25 |
| gpt-6-astra | Economy / unofficial route | ¥1.8 | ¥9 | 2026-09-25 |
| gpt-6-astra | Premium / official route | ¥3 | ¥15 | 2026-09-25 |
Official published price · 2026-09-25 · Price source
Billing evidence
1 privacy-reviewed ten-request session(s) are matched to the tested-key group, token totals and account ledger. These results apply only to the listed route and date.
| Model / tested-key group | Session date | Observed ten-call charge | Rate calculation | Difference | Rate evidence |
|---|---|---|---|---|---|
| claude-fable-5-1Claude Max external route | 2026-09-18 | $0.063594 | $0.063594 | $0 | Later same-group quote matchedThe current same-group quote reviewed after the session exactly matches the ten-call log; it does not prove that the historical rate never changed. |
Review all account-meter sessions and remaining billing gaps · Review failure and retest timeline
What the evidence means
Paid samples cover multiple GPT and Claude models through both Responses and Claude Code compatible protocols.
The Claude Fable 5.1 ten-call ledger is reconciled to the same-group rate; that result does not extend to other models or groups.
Users who value broad model choice and will verify success rate, latency and billing with a small recharge.
Billing evidence for other models and groups, model identity controls, cache pricing and sustained concurrency tests.
Who, how and commercial disclosure
Website probes, provider-published prices and paid model calls are stored as separate evidence types. Failed calls remain in the sample.
API request costs are paid from AI Relay Check test accounts. Account credit can include our recharge and provider promotional credit; those sources are not yet separated in every record.
This provider has a dedicated registration offer or partner entry. AI Relay Check may receive referral benefits; commercial relationships do not change samples or ranking weights.
Corrections and appeals: airelaycheck@163.com · Open raw evidence JSON
Answer summary
- Latest test date
- Provider covered
- H API
- Models / protocols
- claude-fable-5-1 / claude-opus-5 / claude-sonnet-5 +4 more · ClaudeCode / Responses
- Successful / total samples
- 28 / 36
- Price / billing verified
- price yes · billing yes
- Still unverified
- Billing evidence for other models and groups, model identity controls, cache pricing and sustained concurrency tests.
- One-sentence conclusion
- H API has 28/36 successful recorded requests; use the exact model, protocol, route and billing evidence rather than this aggregate alone.