Observation window and scope
2026-10-02T07:08:04.951Z to 2026-10-03T07:05:00.000Z (UTC, start inclusive, end exclusive); exactly 23.948624722222224 hours. First scheduled slot: 2026-10-02T07:15:00.000Z. The observation has ended.
Regions: Seoul, N. Virginia, Oregon. Providers: Anthropic, Google, OpenAI. Monitored models (3): Claude Haiku 4.5, Gemini 3.5 Flash-Lite, GPT-6 Luna.
Sample and error accounting
252 attempts · 245 successful · 7 non-successes.
- Health: 216 attempts, 209 successful, 7 non-successes.
- Speed: 36 attempts, 36 successful, 0 non-successes.
Expected scheduled slots: 252. Duplicates: 0. Missing scheduled slots: 0. Every recorded attempt is included in reliability accounting. Invalid identity, failed and missing timing records are excluded only from timing summaries. Missing scheduled slots are not successful observations.
No retries were made. Prior smoke records are excluded from this report.
| samples | 252 |
|---|---|
| successful | 245 |
| failed | 7 |
| firstAttemptFailures | 7 |
| duplicates | 0 |
| expectedSlots | 252 |
| missingScheduledSlots | 0 |
| excludedTiming | 7 |
| reservedMicros | 233352 |
Failure inventory
- http-error: 1
- incomplete: 6
- success: 245
| Scheduled slot (UTC) | Provider / model | Region / profile | Outcome / HTTP |
|---|---|---|---|
| 2026-10-03T00:15:00.000Z | Google / Gemini 3.5 Flash-Lite | N. Virginia / health | http-error / 503 |
| 2026-10-03T05:15:00.000Z | OpenAI / GPT-6 Luna | Seoul / health | incomplete / 200 |
| 2026-10-03T06:15:00.000Z | OpenAI / GPT-6 Luna | Seoul / health | incomplete / 200 |
| 2026-10-03T05:15:00.000Z | OpenAI / GPT-6 Luna | N. Virginia / health | incomplete / 200 |
| 2026-10-03T06:15:00.000Z | OpenAI / GPT-6 Luna | N. Virginia / health | incomplete / 200 |
| 2026-10-03T05:15:00.000Z | OpenAI / GPT-6 Luna | Oregon / health | incomplete / 200 |
| 2026-10-03T06:15:00.000Z | OpenAI / GPT-6 Luna | Oregon / health | incomplete / 200 |
OpenAI: six HTTP 200 incomplete health records at two hourly slots across all three regions. First stream events were observed, but no visible text, resolved model or usage was retained. This is an unresolved cross-region synchronized incomplete pattern. Existing normalization merges failed/incomplete terminal states and discards their details; the original stream payloads were not retained, so the cause cannot be established from this dataset.
Google: one health HTTP 503 in N. Virginia, attributed to provider-service by the HTTP classification. Anthropic: 84/84 observed successful. These counts do not establish consumer app outages.
Regional coverage
- Seoul: 84 recorded / 84 expected scheduled slots
- N. Virginia: 84 recorded / 84 expected scheduled slots
- Oregon: 84 recorded / 84 expected scheduled slots
Timing methodology
TTFT is elapsed time from request dispatch until the first nonempty user-visible text delta, measured with a monotonic clock. Health is a short availability request; speed uses a longer fixed request. They are separate experiments. No regional distributions are merged into a global latency.
p50 uses the existing lower empirical median: sorted value at index floor((n−1)/2), rather than averaging the middle pair. Only successful valid resolved-model timings qualify. p95 is withheld below 30 comparable timing samples; 22 of 22 comparison groups are below that threshold. Min/max and exact comparison settings remain available below.
Health TTFT
| Provider / model | Seoul | N. Virginia | Oregon | Attempts | Non-successes |
|---|---|---|---|---|---|
| Anthropic Claude Haiku 4.5 | 1000.2 ms 24/24 timing; 0 non-success | 466.9 ms 24/24 timing; 0 non-success | 497.8 ms 24/24 timing; 0 non-success | 72 | 0 |
| Google Gemini 3.5 Flash-Lite | 1101.3 ms 24/24 timing; 0 non-success | 901.3 ms 23/24 timing; 1 non-success | 510.8 ms 24/24 timing; 0 non-success | 72 | 1 |
| OpenAI GPT-6 Luna | 1496.2 ms 22/24 timing; 2 non-success | 1482.0 ms 22/24 timing; 2 non-success | 2059.4 ms 22/24 timing; 2 non-success | 72 | 6 |
Speed TTFT
| Provider / model | Seoul | N. Virginia | Oregon | Attempts | Non-successes |
|---|---|---|---|---|---|
| Anthropic Claude Haiku 4.5 | 620.0 ms 4/4 timing; 0 non-success | 941.0 ms 4/4 timing; 0 non-success | 981.9 ms 4/4 timing; 0 non-success | 12 | 0 |
| Google Gemini 3.5 Flash-Lite | 626.0 ms 4/4 timing; 0 non-success | 468.6 ms 4/4 timing; 0 non-success | 452.7 ms 4/4 timing; 0 non-success | 12 | 0 |
| OpenAI GPT-6 Luna | 1044.4 ms 4/4 timing; 0 non-success | 1879.5 ms 4/4 timing; 0 non-success | 853.3 ms 4/4 timing; 0 non-success | 12 | 0 |
Exact comparison groups, samples and settings
claude-haiku-4-5-20251001 · ap-northeast-2 · health
- provider
- anthropic
- model
- claude-haiku-4-5-20251001
- revision
- catalog-2026-10-02:claude-haiku-4-5-20251001
- region
- ap-northeast-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- disabled
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 24; successful: 24; failed: 0; valid timing samples: 24.
Observed p50 TTFT: 1000.2 ms. p95 TTFT: Withheld. Minimum 926.6 ms; maximum 1439.2 ms.
claude-haiku-4-5-20251001 · ap-northeast-2 · speed
- provider
- anthropic
- model
- claude-haiku-4-5-20251001
- revision
- catalog-2026-10-02:claude-haiku-4-5-20251001
- region
- ap-northeast-2
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- disabled
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 620.0 ms. p95 TTFT: Withheld. Minimum 588.1 ms; maximum 637.2 ms.
claude-haiku-4-5-20251001 · us-east-1 · health
- provider
- anthropic
- model
- claude-haiku-4-5-20251001
- revision
- catalog-2026-10-02:claude-haiku-4-5-20251001
- region
- us-east-1
- profile
- health
- promptVersion
- health-v1
- reasoning
- disabled
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 24; successful: 24; failed: 0; valid timing samples: 24.
Observed p50 TTFT: 466.9 ms. p95 TTFT: Withheld. Minimum 402.5 ms; maximum 546.4 ms.
claude-haiku-4-5-20251001 · us-east-1 · speed
- provider
- anthropic
- model
- claude-haiku-4-5-20251001
- revision
- catalog-2026-10-02:claude-haiku-4-5-20251001
- region
- us-east-1
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- disabled
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 941.0 ms. p95 TTFT: Withheld. Minimum 919.2 ms; maximum 1038.8 ms.
claude-haiku-4-5-20251001 · us-west-2 · health
- provider
- anthropic
- model
- claude-haiku-4-5-20251001
- revision
- catalog-2026-10-02:claude-haiku-4-5-20251001
- region
- us-west-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- disabled
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 24; successful: 24; failed: 0; valid timing samples: 24.
Observed p50 TTFT: 497.8 ms. p95 TTFT: Withheld. Minimum 441.3 ms; maximum 584.0 ms.
claude-haiku-4-5-20251001 · us-west-2 · speed
- provider
- anthropic
- model
- claude-haiku-4-5-20251001
- revision
- catalog-2026-10-02:claude-haiku-4-5-20251001
- region
- us-west-2
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- disabled
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 981.9 ms. p95 TTFT: Withheld. Minimum 938.9 ms; maximum 1009.3 ms.
gemini-3.5-flash-lite · ap-northeast-2 · health
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:gemini-3.5-flash-lite
- region
- ap-northeast-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- minimal
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 24; successful: 24; failed: 0; valid timing samples: 24.
Observed p50 TTFT: 1101.3 ms. p95 TTFT: Withheld. Minimum 959.2 ms; maximum 7600.6 ms.
gemini-3.5-flash-lite · ap-northeast-2 · speed
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:gemini-3.5-flash-lite
- region
- ap-northeast-2
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- minimal
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 626.0 ms. p95 TTFT: Withheld. Minimum 573.7 ms; maximum 672.1 ms.
gemini-3.5-flash-lite · us-east-1 · health
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:gemini-3.5-flash-lite
- region
- us-east-1
- profile
- health
- promptVersion
- health-v1
- reasoning
- minimal
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 23; successful: 23; failed: 0; valid timing samples: 23.
Observed p50 TTFT: 901.3 ms. p95 TTFT: Withheld. Minimum 720.8 ms; maximum 5579.1 ms.
gemini-3.5-flash-lite · us-east-1 · speed
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:gemini-3.5-flash-lite
- region
- us-east-1
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- minimal
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 468.6 ms. p95 TTFT: Withheld. Minimum 378.5 ms; maximum 525.3 ms.
gemini-3.5-flash-lite · us-west-2 · health
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:gemini-3.5-flash-lite
- region
- us-west-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- minimal
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 24; successful: 24; failed: 0; valid timing samples: 24.
Observed p50 TTFT: 510.8 ms. p95 TTFT: Withheld. Minimum 358.1 ms; maximum 16857.0 ms.
gemini-3.5-flash-lite · us-west-2 · speed
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:gemini-3.5-flash-lite
- region
- us-west-2
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- minimal
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 452.7 ms. p95 TTFT: Withheld. Minimum 383.3 ms; maximum 535.7 ms.
gemini-3.5-flash-lite · us-east-1 · health
- provider
- model
- gemini-3.5-flash-lite
- revision
- catalog-2026-10-02:unresolved
- region
- us-east-1
- profile
- health
- promptVersion
- health-v1
- reasoning
- minimal
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 1; successful: 0; failed: 1; valid timing samples: 0.
Observed p50 TTFT: Withheld. p95 TTFT: Withheld.
gpt-6-luna · ap-northeast-2 · health
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:gpt-6-luna
- region
- ap-northeast-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- none
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 22; successful: 22; failed: 0; valid timing samples: 22.
Observed p50 TTFT: 1496.2 ms. p95 TTFT: Withheld. Minimum 978.7 ms; maximum 2227.1 ms.
gpt-6-luna · ap-northeast-2 · speed
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:gpt-6-luna
- region
- ap-northeast-2
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- none
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 1044.4 ms. p95 TTFT: Withheld. Minimum 770.7 ms; maximum 1294.3 ms.
gpt-6-luna · us-east-1 · health
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:gpt-6-luna
- region
- us-east-1
- profile
- health
- promptVersion
- health-v1
- reasoning
- none
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 22; successful: 22; failed: 0; valid timing samples: 22.
Observed p50 TTFT: 1482.0 ms. p95 TTFT: Withheld. Minimum 539.5 ms; maximum 4785.0 ms.
gpt-6-luna · us-east-1 · speed
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:gpt-6-luna
- region
- us-east-1
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- none
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 1879.5 ms. p95 TTFT: Withheld. Minimum 1345.2 ms; maximum 3378.8 ms.
gpt-6-luna · us-west-2 · health
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:gpt-6-luna
- region
- us-west-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- none
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 22; successful: 22; failed: 0; valid timing samples: 22.
Observed p50 TTFT: 2059.4 ms. p95 TTFT: Withheld. Minimum 1598.6 ms; maximum 3126.6 ms.
gpt-6-luna · us-west-2 · speed
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:gpt-6-luna
- region
- us-west-2
- profile
- speed
- promptVersion
- speed-v1
- reasoning
- none
- maxOutputTokens
- 1024
- cachePolicy
- fixed-prompt-provider-managed
Samples: 4; successful: 4; failed: 0; valid timing samples: 4.
Observed p50 TTFT: 853.3 ms. p95 TTFT: Withheld. Minimum 680.1 ms; maximum 1288.8 ms.
gpt-6-luna · ap-northeast-2 · health
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:unresolved
- region
- ap-northeast-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- none
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 2; successful: 0; failed: 2; valid timing samples: 0.
Observed p50 TTFT: Withheld. p95 TTFT: Withheld.
gpt-6-luna · us-east-1 · health
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:unresolved
- region
- us-east-1
- profile
- health
- promptVersion
- health-v1
- reasoning
- none
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 2; successful: 0; failed: 2; valid timing samples: 0.
Observed p50 TTFT: Withheld. p95 TTFT: Withheld.
gpt-6-luna · us-west-2 · health
- provider
- openai
- model
- gpt-6-luna
- revision
- catalog-2026-10-02:unresolved
- region
- us-west-2
- profile
- health
- promptVersion
- health-v1
- reasoning
- none
- maxOutputTokens
- 128
- cachePolicy
- fixed-prompt-provider-managed
Samples: 2; successful: 0; failed: 2; valid timing samples: 0.
Observed p50 TTFT: Withheld. p95 TTFT: Withheld.
Cost: reservations, usage estimate and invoice
Reserved worst-case provider cost: $0.233352 (233352 microUSD). Audited observation ledger reservations: 233352 microUSD. Reservations are conservative budget holds, not provider-billed cost.
Usage-based estimate for 245 records with reliable usage: $0.01639435. Usage is unavailable for 7 records; their $0.001242 reservation remains separate, not assumed to be zero. Known usage estimate plus the unknown-usage reservations is $0.01763635.
Immutable input and billedOutput tokens multiplied by the observation CONTROL v20 price contracts; no rounding per record. OpenAI input uses conservative cache-write ceiling $0.125/MTok. Missing usage is unknown, never zero. This is provider inference cost only; AWS, taxes, credits and account-specific billing adjustments are excluded.
Actual provider invoice amount not independently verified.
Limitations
- Only 23.948624722222224 hours, 3 regions and 3 monitored models.
- Direct API observations do not measure consumer ChatGPT, Claude or Gemini app status, answer quality or provider server location.
- Health and Speed are separate profiles; connection reuse and provider-managed caching can affect timings.
- Throughput is unavailable without reliable visible-token methodology. Stream delta counts are not token counts.
- No seven-day baseline. Best Time and composite score remain off. This dataset does not support winner or general provider performance claims.
- Non-successes are retained. Missing usage and terminal details limit attribution and cost accuracy.
Citation and evidence
- Source
- AI Fast Now independent monitoring
- Measurement method
- Direct provider APIs
- Observation window
- 2026-10-02T07:08:04.951Z–2026-10-03T07:05:00.000Z
- Regions
- ap-northeast-2, us-east-1, us-west-2
- Models
- claude-haiku-4-5-20251001, gemini-3.5-flash-lite, gpt-6-luna
- Samples
- 252 recorded, 245 successful, 7 failed
- Updated
- 2026-10-03T08:10:27.526Z
- Evidence fingerprint
- 1af15003df26dbed4b6c68669ca78226ca6328991b2308c3cee2491bd38051af (audited source dataset; raw records are not publicly distributed)
- Methodology
- AI Fast Now methodology