Skip to content

Model API reliability on Router One

Success rate, rate-limit and server-error share, time to first token and throughput for every listed model with enough traffic to report, aggregated from real calls through Router One over the last 7 and 30 days.

These are observations of production traffic, not a controlled benchmark. Each model serves its own mix of prompt sizes, output lengths and reasoning settings, so the latency columns describe what callers of that model saw on Router One, not how one model's speed compares with another's. Each model's page under /models shows the same aggregates for that model.

Method last revised: ·

Last 7 days

32 listed models, requests from 2026-11-12 to 2026-11-19 (UTC), ordered by request volume
ModelSuccess 2xxRate limit 429Server 5xxAvg. time to first tokenAvg. latencyThroughput (tokens/s)
GPT-5.6 Lunaopenai/gpt-5.6-luna99.98%0.00%0.02%8.56 s2.34 s197.65
GPT-6.1 Solopenai/gpt-6.1-sol99.49%0.04%0.47%10.91 s15.18 s71.41
DeepSeek V4.1 Flashdeepseek-v4.1-flash99.97%0.01%0.02%3.39 s6.78 s262.71
GPT-6 Solopenai/gpt-6-sol99.95%0.00%0.05%10.07 s9.64 s73.58
Claude Haiku 5.5anthropic/claude-haiku-5.599.89%0.00%0.11%5.44 s7.33 s399.35
GPT-5.6 Terraopenai/gpt-5.6-terra99.25%0.00%0.75%14.36 s20.74 s97.32
Claude Sonnet 5.5anthropic/claude-sonnet-5.599.98%0.00%0.02%9.48 s12.37 s231.77
Claude Opus 5.5anthropic/claude-opus-5.599.96%0.00%0.04%12.57 s17.05 s352.13
GPT-5.6 Solopenai/gpt-5.6-sol99.73%0.00%0.27%15.47 s22.02 s84.99
GPT-6.2 Solopenai/gpt-6.2-sol99.57%0.02%0.41%8.69 s15.97 s69.92
GPT-6.1 Terraopenai/gpt-6.1-terra98.91%0.00%1.09%9.77 s16.11 s66.64
GPT-6 Astraopenai/gpt-6-astra99.54%0.00%0.46%19.99 s26.91 s79.92
Claude Sonnet 5anthropic/claude-sonnet-599.60%0.00%0.40%11.7 s17.46 s221.92
Claude Sonnet 4.6 · AWSaws/claude-sonnet-4.696.64%0.00%3.36%—2.24 s—
Grok 4.5grok-4.599.73%0.00%0.27%4.76 s24.87 s56.59
Grok 4.7grok-4.799.63%0.00%0.37%5.94 s19.37 s67.29
Claude Sonnet 4.6anthropic/claude-sonnet-4.699.68%0.06%0.26%8.57 s15.14 s221.16
Claude Haiku 4.5anthropic/claude-haiku-4.597.83%0.00%2.17%6.44 s8.98 s346.42
Claude Fable 5.2anthropic/claude-fable-5.299.92%0.00%0.08%13.83 s26.21 s121.29
DeepSeek V4 Flashdeepseek-v4-flash99.94%0.00%0.06%3 s5.53 s270.17
Grok 4.6grok-4.699.95%0.00%0.05%6.17 s36.63 s67.89
Claude Opus 5anthropic/claude-opus-599.67%0.00%0.33%15.94 s23.28 s381.76
GPT-6.2 Astraopenai/gpt-6.2-astra99.92%0.00%0.08%10.86 s10.96 s70.94
GPT-6.1 Sol · Azureazure/gpt-6.1-sol99.88%0.00%0.12%10.85 s14.38 s87.08
Claude Opus 4.8anthropic/claude-opus-4.899.84%0.00%0.16%16.83 s23.2 s300.92
GPT-6 Astra · Azureazure/gpt-6-astra100.00%0.00%0.00%13.56 s18.03 s101.89
Claude Opus 4.7 (thinking)anthropic/claude-opus-4.7-thinking100.00%0.00%0.00%9.19 s25.35 s152.48
Claude Opus 4.6anthropic/claude-opus-4.699.35%0.00%0.65%14.44 s18.54 s295.75
Grok 4.7 Build Fastgrok-4.7-build-fast100.00%0.00%0.00%3.11 s30.64 s89.18
GPT-6 Sol · Azureazure/gpt-6-sol98.91%0.00%1.09%5.88 s6.41 s211.3
GPT-5.6 Luna · Azureazure/gpt-5.6-luna96.84%0.00%3.16%12.43 s12.89 s190.17
Claude Sonnet 4.6 (thinking)anthropic/claude-sonnet-4.6-thinking100.00%0.00%0.00%15.39 s25.72 s293.84

Last 30 days

43 listed models, requests from 2026-10-20 to 2026-11-19 (UTC), ordered by request volume
ModelSuccess 2xxRate limit 429Server 5xxAvg. time to first tokenAvg. latencyThroughput (tokens/s)
GPT-6.1 Solopenai/gpt-6.1-sol99.64%0.01%0.35%12.07 s17.28 s153.16
GPT-5.6 Luna · Azureazure/gpt-5.6-luna99.91%0.03%0.05%6.16 s7.6 s270.02
DeepSeek V4.1 Flashdeepseek-v4.1-flash99.97%0.00%0.03%3.69 s7.2 s265.95
GPT-5.6 Terraopenai/gpt-5.6-terra99.46%0.00%0.54%14.79 s19.09 s117.74
GPT-5.6 Lunaopenai/gpt-5.6-luna99.94%0.00%0.06%6.24 s2.47 s201.24
GPT-6 Solopenai/gpt-6-sol99.92%0.00%0.08%10.23 s12.81 s183.4
Claude Sonnet 5.5anthropic/claude-sonnet-5.599.87%0.01%0.12%7.31 s12.21 s180.9
Claude Opus 5.5anthropic/claude-opus-5.599.76%0.02%0.22%12.61 s16.73 s401.65
GPT-6 Astraopenai/gpt-6-astra99.69%0.00%0.31%13.66 s21.68 s132.79
Claude Haiku 5.5anthropic/claude-haiku-5.599.42%0.01%0.57%10.79 s13.2 s218.08
GPT-5.6 Solopenai/gpt-5.6-sol99.60%0.00%0.40%14.86 s20.53 s110.46
GPT-6.1 Terraopenai/gpt-6.1-terra98.89%0.01%1.11%9.82 s13.25 s87.91
Claude Sonnet 5anthropic/claude-sonnet-599.72%0.00%0.28%13 s17.56 s191.94
Grok 4.5grok-4.599.09%0.01%0.90%4.97 s23.32 s59.47
Grok 4.7grok-4.798.62%0.01%1.37%5.17 s20.15 s72.78
Claude Sonnet 4.6anthropic/claude-sonnet-4.699.67%0.02%0.30%9.29 s14.93 s181.43
Claude Sonnet 4.6 · AWSaws/claude-sonnet-4.698.55%0.00%1.45%—2.17 s—
Claude Haiku 4.5anthropic/claude-haiku-4.599.19%0.00%0.81%4.69 s7.57 s234.86
GPT-6.2 Solopenai/gpt-6.2-sol99.57%0.02%0.41%8.69 s15.97 s69.92
DeepSeek V4 Flashdeepseek-v4-flash99.94%0.00%0.06%2.61 s4.88 s255.23
Grok 4.6grok-4.699.12%0.01%0.87%6.17 s35.72 s67.95
Claude Fable 5.2anthropic/claude-fable-5.299.89%0.00%0.11%13.94 s26.5 s114.05
Claude Opus 5anthropic/claude-opus-599.48%0.00%0.52%15.11 s22.95 s249.52
Claude Sonnet 4.6 (thinking)anthropic/claude-sonnet-4.6-thinking99.88%0.00%0.12%8.69 s19.64 s160.94
Claude Fable 5.1anthropic/claude-fable-5.199.91%0.00%0.09%19.8 s25.99 s183.71
Codex Auto Reviewcodex-auto-review98.89%0.00%1.11%8.61 s12.28 s76.8
Claude Opus 4.6anthropic/claude-opus-4.699.14%0.00%0.86%9.6 s14.01 s191.75
Claude Opus 4.8anthropic/claude-opus-4.899.64%0.00%0.36%17.06 s33 s255.84
GPT-6.1 Sol · Azureazure/gpt-6.1-sol98.70%0.57%0.73%12.04 s16.43 s84.6
Grok 4.7 Build Fastgrok-4.7-build-fast96.33%0.00%3.67%3.44 s18.33 s118.3
GPT-6.2 Astraopenai/gpt-6.2-astra99.92%0.00%0.08%10.86 s10.96 s70.94
GPT-6 Sol · Azureazure/gpt-6-sol97.94%0.00%2.06%9.03 s9.77 s289.43
GPT-6 Astra · Azureazure/gpt-6-astra99.00%0.14%0.86%13.01 s18.66 s94.92
GPT-5.5 · Azureazure/gpt-5.592.48%0.16%7.36%8.58 s13.18 s139.6
Claude Opus 4.7 (thinking)anthropic/claude-opus-4.7-thinking98.79%0.61%0.61%9.03 s25.29 s151.63
Claude Haiku 4.5 · AWSaws/claude-haiku-4.590.20%0.00%9.80%—1.85 s—
Claude Fable 5anthropic/claude-fable-599.64%0.00%0.36%13.94 s14.36 s142.03
Claude Opus 4.6 (thinking)anthropic/claude-opus-4.6-thinking75.72%0.00%24.28%—9.36 s—
GPT-5.6 Terra · Azureazure/gpt-5.6-terra98.50%0.00%1.50%—16.41 s—
GPT-5.6 Sol · Azureazure/gpt-5.6-sol100.00%0.00%0.00%9.39 s10.76 s154.71
Claude Sonnet 5 · AWSaws/claude-sonnet-586.44%1.13%12.43%—11.89 s—
Claude Sonnet 4.5 · AWSaws/claude-sonnet-4.588.89%3.47%7.64%—10.58 s—
Claude Opus 4.6 · AWSaws/claude-opus-4.698.35%0.00%1.65%—10.87 s—

Sales channels beside the default channel, last 30 days

Ids prefixed azure/, aws/, vertex/ or aistudio/ are separately priced sales channels of a base model, each with its own routing pool. The rows below put a channel id next to the default-channel id of the same model where both cleared the sample floor. The two serve different callers and workloads, so read the difference as what each channel's traffic saw, not as a like-for-like test.

ModelChannelSuccess 2xxRate limit 429Server 5xxAvg. time to first tokenAvg. latencyThroughput (tokens/s)
GPT-6.1 Solopenai/gpt-6.1-solDefault99.64%0.01%0.35%12.07 s17.28 s153.16
GPT-6.1 Sol · Azureazure/gpt-6.1-solAzure98.70%0.57%0.73%12.04 s16.43 s84.6
GPT-5.6 Terraopenai/gpt-5.6-terraDefault99.46%0.00%0.54%14.79 s19.09 s117.74
GPT-5.6 Terra · Azureazure/gpt-5.6-terraAzure98.50%0.00%1.50%—16.41 s—
GPT-5.6 Lunaopenai/gpt-5.6-lunaDefault99.94%0.00%0.06%6.24 s2.47 s201.24
GPT-5.6 Luna · Azureazure/gpt-5.6-lunaAzure99.91%0.03%0.05%6.16 s7.6 s270.02
GPT-6 Solopenai/gpt-6-solDefault99.92%0.00%0.08%10.23 s12.81 s183.4
GPT-6 Sol · Azureazure/gpt-6-solAzure97.94%0.00%2.06%9.03 s9.77 s289.43
GPT-6 Astraopenai/gpt-6-astraDefault99.69%0.00%0.31%13.66 s21.68 s132.79
GPT-6 Astra · Azureazure/gpt-6-astraAzure99.00%0.14%0.86%13.01 s18.66 s94.92
GPT-5.6 Solopenai/gpt-5.6-solDefault99.60%0.00%0.40%14.86 s20.53 s110.46
GPT-5.6 Sol · Azureazure/gpt-5.6-solAzure100.00%0.00%0.00%9.39 s10.76 s154.71
Claude Sonnet 5anthropic/claude-sonnet-5Default99.72%0.00%0.28%13 s17.56 s191.94
Claude Sonnet 5 · AWSaws/claude-sonnet-5AWS86.44%1.13%12.43%—11.89 s—
Claude Sonnet 4.6anthropic/claude-sonnet-4.6Default99.67%0.02%0.30%9.29 s14.93 s181.43
Claude Sonnet 4.6 · AWSaws/claude-sonnet-4.6AWS98.55%0.00%1.45%—2.17 s—
Claude Haiku 4.5anthropic/claude-haiku-4.5Default99.19%0.00%0.81%4.69 s7.57 s234.86
Claude Haiku 4.5 · AWSaws/claude-haiku-4.5AWS90.20%0.00%9.80%—1.85 s—
Claude Opus 4.6anthropic/claude-opus-4.6Default99.14%0.00%0.86%9.6 s14.01 s191.75
Claude Opus 4.6 · AWSaws/claude-opus-4.6AWS98.35%0.00%1.65%—10.87 s—

How to read the columns

Success 2xx, rate limit 429, server 5xx
The share of requests whose final status was a 2xx, a 429 or a 5xx. Only requests that ended in one of those three are counted; a request refused for a malformed body or an empty balance is in no column.
Avg. time to first token
The mean over successful streaming requests that recorded a first token. It grows with prompt size and with how long the model works before it starts to stream.
Avg. latency
The mean time to the complete response over successful requests. A model that is asked for long outputs shows a longer average.
Throughput
Output tokens per second of generation, over successful streaming requests that recorded both a token count and a generation time.
Sample floor
A model appears in a window only with at least 100 eligible requests from at least 5 independent principals in that window. A dash in a cell means the window holds no request of the kind that column is computed from.
Row order
Rows are ordered by request volume in the window. Nothing on this page ranks one model above another.

What this page is not

Not a speed comparison between models
The requests behind each row are whatever that model's callers sent. A model used mostly for long agentic sessions and one used for short classification calls will differ on every latency column for reasons that have nothing to do with the model.
Not a service-level commitment
Contractual availability is documented on /sla, and current incidents are on /status. This page reports what happened, window by window.
Not your own traffic
The aggregates cover all callers. Dashboard → Logs records the status, latency and token counts of each of your own requests.

FAQ

How often is this data updated?

This page re-reads the gateway's aggregates within the hour. The line under the heading shows when the page last read them, and each table states the dates its window covers.

Why is a model I use missing from the table?

A model is listed only when it is in the current catalog and had at least 100 eligible requests from at least 5 independent principals in the window. A model below that floor in the last 7 days can still appear in the 30-day table.

Why do the 7-day and 30-day rows for one model differ?

They cover different requests. A short incident or a change in what callers send moves the 7-day row quickly and the 30-day row slowly, so read the two together.

Does a 429 in this table come from my own rate limit?

The 429 column counts requests whose final status was 429 across all callers. Per-key limits and budgets are set by each account, and your own requests and their statuses are in Dashboard → Logs.

Related

Call any of these models with one API key