Independent benchmarks by Artificial Analysis

The fastest provider on every model we’re measured on.

Artificial Analysis benchmarks inference providers side by side, on the same open models and the same workloads, on their public API.

Headline gains.

8.2×
faster than Kimi’s own APIKimi K3.1 Pro · time per task · less time
9.2×
cheaper than Baseten (Fast)GLM-5.3 Max · cost per task · lower cost
4.2×
faster than DatabricksKimi K3.1 Pro · time per task · less time
13.8×
cheaper than SambaNovaGLM-5.3 Low · cost per task · lower cost
3.2×
faster than NebiusGLM-5.3 Max · time per task · less time
5.2×
cheaper than FireworksGLM-5.3 Max · cost per task · lower cost

Every result.

Grouped by model, each with the chart as published by Artificial Analysis.

Kimi K3.1 Pro

Output speed

265 tokens per second

#1 of 19 endpoints on output speed. 8.8× Kimi's own API.

Best gain 8.8×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: Kimi K3.1 Pro output speed, Systalyze first at 265 tokens per second
Kimi K3.1 Pro · Output speed. Screenshot from Artificial Analysis.
Speed by workload

243 tokens per second in multi-turn sessions

#1 of 19 providers and fastest on every workload. In multi-turn, 1.4× Baseten and 7.5× Kimi's own API.

Best gain 7.5×

Output speed by workload. End-to-end output speed for four workloads: 1k, 10k and 100k-token prompts and multi-turn agent sessions. The table shows multi-turn. Higher is better.

Artificial Analysis chart: Kimi K3.1 Pro output speed by workload, Systalyze first at 243 tokens per second in multi-turn
Kimi K3.1 Pro · Speed by workload. Screenshot from Artificial Analysis.
Error rate

0.0% errors

Zero failed requests across every single-request and multi-turn agent workload measured.

Error rate. The share of requests that failed with an error, a rate limit or a timeout, across every single-request and multi-turn agent workload measured. A fast endpoint is only useful if the answer arrives. Lower is better.

Artificial Analysis chart: Kimi K3.1 Pro endpoint error rate, Systalyze at 0.0%
Kimi K3.1 Pro · Error rate. Screenshot from Artificial Analysis.
Time per task

4.6 minutes per task

#1 of 19 providers. Kimi's own API takes 37.9 minutes, 8.2× longer.

Best gain 8.2×

Time per task. How long it takes to finish one full task from the Artificial Analysis Intelligence Index, end to end: coding, data analysis, workflow automation and more. This is closest to what a user actually feels. Lower is better.

Artificial Analysis chart: Kimi K3.1 Pro time per task, Systalyze first at 4.6 minutes per task
Kimi K3.1 Pro · Time per task. Screenshot from Artificial Analysis.

GLM-5.3 Max

Time per task

4.8 minutes per task

#1 of 29 providers. 1.7× faster than Baseten (Fast) and 5.5× faster than Z.ai's own API.

Best gain 5.5×

Time per task. How long it takes to finish one full task from the Artificial Analysis Intelligence Index, end to end: coding, data analysis, workflow automation and more. This is closest to what a user actually feels. Lower is better.

Artificial Analysis chart: GLM-5.3 Max, Systalyze first at 4.8 minutes per task
GLM-5.3 Max · Time per task. Screenshot from Artificial Analysis.
Output speed

701 tokens per second

#1 of 27 providers on raw output speed, and 8.7× Z.ai's own API.

Best gain 8.6×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: GLM-5.3 Max · speed, Systalyze first at 701 tokens per second
GLM-5.3 Max · Output speed. Screenshot from Artificial Analysis.
Cost per task

$1.04 per task

Cheapest of 30 endpoints: 5.2× cheaper than Fireworks and 9.2× cheaper than Baseten (Fast).

Best gain 9.1×

Cost per task. What it costs to finish one Intelligence Index task, counting every token the model reads and writes. This is what you actually pay. Lower is better.

Artificial Analysis chart: GLM-5.3 Max cost per task, Systalyze cheapest at $1.04 per task
GLM-5.3 Max · Cost per task. Screenshot from Artificial Analysis.
Cost vs time per task

Fastest and cheapest of 29 providers.

#1 of 29 providers on both time and cost per task. About 1.6× faster per task than Baseten (FAST) and 5.4× faster than Z.ai's own API, at the lowest cost per task in the chart.

Cost per task vs time per task. Artificial Analysis plots the weighted average cost in USD to run one Intelligence Index task against the minutes each task takes (output tokens divided by end-to-end output speed). Lower is better on both axes.

Artificial Analysis chart: GLM-5.3 Max cost per task vs time per task, Systalyze fastest and cheapest at about 5.1 minutes and about $1 per task
GLM-5.3 Max · Cost vs time per task. Screenshot from Artificial Analysis.
Output speed vs cost per task

Fastest and cheapest of 29 providers.

#1 of 29 providers on both output speed and cost per task. About 1.6× the speed of Baseten (FAST) and 5.3× the speed of Z.ai's own API, at the lowest cost per task in the chart.

Output speed vs cost per task. Artificial Analysis plots end-to-end output speed (output tokens divided by total request time, including time to first token) against the weighted average cost in USD per Intelligence Index task. Higher speed and lower cost are better.

Artificial Analysis chart: GLM-5.3 Max output speed vs cost per task, Systalyze first at about 385 tokens per second and about $1.0 per task
GLM-5.3 Max · Output speed vs cost per task. Screenshot from Artificial Analysis.
Output speed by workload

289 tokens per second in multi-turn sessions

#1 of 29 providers and fastest on every workload. In multi-turn, 1.3× Baseten (FAST) and 4.9× Z.ai's own API.

Best gain 4.9×

Output speed by workload. End-to-end output speed for four workloads: 1k, 10k and 100k-token prompts and multi-turn agent sessions. The table shows multi-turn. Higher is better.

Artificial Analysis chart: GLM-5.3 Max output speed by workload, Systalyze first at 289 tokens per second in multi-turn
GLM-5.3 Max · Output speed by workload. Screenshot from Artificial Analysis.

GLM-5.3 Low

Time per task

2.6 minutes per task

#1 of 11 providers. Z.ai's own API takes 15 minutes, 5.8× longer.

Best gain 5.7×

Time per task. How long it takes to finish one full task from the Artificial Analysis Intelligence Index, end to end: coding, data analysis, workflow automation and more. This is closest to what a user actually feels. Lower is better.

Artificial Analysis chart: GLM-5.3 Low, Systalyze first at 2.6 minutes per task
GLM-5.3 Low · Time per task. Screenshot from Artificial Analysis.
Cost per task

$0.33 per task

Cheapest of 11 endpoints: 4.7× cheaper than Z.ai's own API and 13.8× cheaper than SambaNova.

Best gain 13.7×

Cost per task. What it costs to finish one Intelligence Index task, counting every token the model reads and writes. This is what you actually pay. Lower is better.

Artificial Analysis chart: GLM-5.3 Low cost per task, Systalyze cheapest at $0.33 per task
GLM-5.3 Low · Cost per task. Screenshot from Artificial Analysis.
Speed vs cost

Fast and cheap. Nobody else gets both.

~330 tokens per second at $0.33 per task: the best endpoint in the most attractive quadrant, fast and cheap at the same time.

Best gain 6.0×

Speed vs cost. Every provider plotted by output speed and by cost per task. The top-left corner, fast and cheap at the same time, is the most attractive quadrant.

Artificial Analysis chart: GLM-5.3 Low output speed vs cost per task, Systalyze fastest and cheapest
GLM-5.3 Low · Speed vs cost. Screenshot from Artificial Analysis.

Gemma 4

Output speed

450 tokens per second

#1 of 7 providers on output speed: 3.5× the next fastest provider and 9.6× Google AI Studio.

Best gain 9.5×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: Gemma 4, Systalyze first at 450 tokens per second
Gemma 4 · Output speed. Screenshot from Artificial Analysis.
Speed by workload

Up to 403 tokens per second, at every prompt length

Fastest on every workload: 379 t/s with 1k-token prompts, 403 with 10k, 296 with 100k and 336 in multi-turn sessions.

Best gain 14.6×

Speed by workload. Output speed measured separately with 1k, 10k and 100k-token prompts and in multi-turn agent sessions. It shows whether a provider stays fast as prompts get longer and conversations go on. Higher is better.

Artificial Analysis chart: Gemma 4 output speed by workload, Systalyze fastest at every workload
Gemma 4 · Speed by workload. Screenshot from Artificial Analysis.
Time to first token

3.69 seconds to first answer token

About half the wait of the next fastest provider, and 6.5× faster than Google AI Studio.

Best gain 6.5×

Time to first token. How long a user waits before the answer starts to appear, after any reasoning. This is what makes an app feel responsive. Lower is better.

Artificial Analysis chart: Gemma 4 time to first answer token, Systalyze first at 3.69 seconds
Gemma 4 · Time to first token. Screenshot from Artificial Analysis.
Error rate

0.0% errors

The only endpoint with zero failed requests, out of 7.

Error rate. The share of requests that failed with an error, a rate limit or a timeout, across every single-request and multi-turn agent workload measured. A fast endpoint is only useful if the answer arrives. Lower is better.

Artificial Analysis chart: Gemma 4 endpoint error rate, Systalyze at 0.0%
Gemma 4 · Error rate. Screenshot from Artificial Analysis.
Speed, non-reasoning

441 tokens per second

3.0× the next fastest provider, and 9.2× CoreWeave.

Best gain 9.1×

Output speed, non-reasoning. Tokens per second with the model's reasoning turned off, the mode used for quick chat and simple tool calls. Higher is better.

Artificial Analysis chart: Gemma 4 · non-reasoning, Systalyze first at 441 tokens per second
Gemma 4 · Speed, non-reasoning. Screenshot from Artificial Analysis.
Blended price

$0.07 per 1M tokens, blended

#1 of 8 providers on blended price. 1.2× cheaper than CoreWeave and 2.2× cheaper than GMI.

Best gain 2.2×

Blended price. The price per 1M tokens, blended 7:2:1 across cached input, input and output tokens. Lower is better.

Artificial Analysis chart: Gemma 4 blended price, Systalyze first at $0.07
Gemma 4 · Blended price. Screenshot from Artificial Analysis.
End-to-end response time

6.6 seconds end to end

#1 of 7 providers on end-to-end response time. 7.9× faster than CoreWeave and 8.2× faster than Google AI Studio.

Best gain 8.2×

End-to-end response time. Seconds to receive 500 output tokens from a 10,000-token prompt, including input processing and the model's thinking time. Lower is better.

Artificial Analysis chart: Gemma 4 end-to-end response time, Systalyze first at 6.6 s
Gemma 4 · End-to-end response time. Screenshot from Artificial Analysis.
End-to-end response time vs price

Fastest and cheapest of 6 providers.

Systalyze has the lowest end-to-end response time and the lowest blended price of the 6 providers in the chart. CoreWeave takes about 52 seconds at a higher price.

End-to-end response time vs price. Each provider is plotted by seconds to output 500 tokens against blended price per 1M tokens. Lower and further left is better.

Artificial Analysis chart: Gemma 4 response time vs price, Systalyze first at about 6.6 s and $0.07 per 1M tokens
Gemma 4 · End-to-end response time vs price. Screenshot from Artificial Analysis.
Output speed vs price

Fastest and cheapest of 6 providers.

Systalyze has the highest output speed and the lowest blended price of the 6 providers in the chart. CoreWeave runs at about 50 tokens per second at a higher price.

Output speed vs price. Each provider is plotted by output tokens per second against blended price per 1M tokens, blended 7:2:1 cache to input to output. Higher and further left is better.

Artificial Analysis chart: Gemma 4 speed vs price, Systalyze first at about 450 t/s and $0.07 per 1M tokens
Gemma 4 · Output speed vs price. Screenshot from Artificial Analysis.
Output speed variance

450 tokens per second, median

#1 of 7 providers on median output speed. 9.1× faster than CoreWeave and 9.5× faster than Google AI Studio.

Best gain 9.5×

Output speed variance. The spread of output tokens per second across many requests with a 10,000-token prompt, shown by percentile; the value is the median. Higher is better.

Artificial Analysis chart: Gemma 4 speed variance, Systalyze first at 450 t/s
Gemma 4 · Output speed variance. Screenshot from Artificial Analysis.

GLM-5.2

Output speed, Max

538 tokens per second

#1 of 16 providers: 1.6× Baseten (Fast) and 5.5× Z.ai's own API.

Best gain 5.4×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: GLM-5.2 Max, Systalyze first at 538 tokens per second
GLM-5.2 · Output speed, Max. Screenshot from Artificial Analysis.
Speed, non-reasoning

459 tokens per second

1.4× faster than Baseten (Fast).

Best gain 3.9×

Output speed, non-reasoning. Tokens per second with the model's reasoning turned off, the mode used for quick chat and simple tool calls. Higher is better.

Artificial Analysis chart: GLM-5.2 · non-reasoning, Systalyze first at 459 tokens per second
GLM-5.2 · Speed, non-reasoning. Screenshot from Artificial Analysis.

DeepSeek V4 Pro

Output speed

277 tokens per second

#1 of 11 providers on output speed: 1.4× Nebius and 4.0× DeepSeek's own API.

Best gain 4.0×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: DeepSeek V4 Pro, Systalyze first at 277 tokens per second
DeepSeek V4 Pro · Output speed. Screenshot from Artificial Analysis.
Blended price

$0.33 per 1M tokens, blended

#1 of 11 providers on blended price. 2.0× cheaper than DeepSeek's own API and 4.7× cheaper than Nebius.

Best gain 4.7×

Blended price. The cost per 1M tokens, blending cached input, input and output prices at a 7:2:1 ratio. Lower is better.

Artificial Analysis chart: DeepSeek V4 Pro blended price, Systalyze first at $0.33
DeepSeek V4 Pro · Blended price. Screenshot from Artificial Analysis.
Time to first answer token

8.6 seconds to first answer token

#1 of 11 providers on time to first answer token. 1.4× faster than Nebius and 3.5× faster than DeepSeek's own API.

Best gain 4.1×

Time to first answer token. Seconds from sending a 10,000-token prompt until the first answer token arrives, including input processing and the model's thinking time. Lower is better.

Artificial Analysis chart: DeepSeek V4 Pro time to first answer token, Systalyze first at 8.6 s
DeepSeek V4 Pro · Time to first answer token. Screenshot from Artificial Analysis.
End-to-end response time

10.4 seconds end to end

#1 of 11 providers on end-to-end response time. 1.4× faster than Nebius and 3.6× faster than DeepSeek's own API.

Best gain 4.2×

End-to-end response time. Seconds to receive a 500-token answer to a 10,000-token prompt, including input processing and the model's thinking time. Lower is better.

Artificial Analysis chart: DeepSeek V4 Pro end-to-end response time, Systalyze first at 10.4 s
DeepSeek V4 Pro · End-to-end response time. Screenshot from Artificial Analysis.
Output speed vs price

Fastest and cheapest of 11 providers.

Systalyze is the only provider in the most attractive quadrant. 1.4× faster than Nebius at 4.7× lower price, and 4.0× faster than DeepSeek's own API at 2.0× lower price.

Output speed vs price. Output tokens per second plotted against blended price per 1M tokens (7:2:1 cache to input to output). Higher and further left is better.

Artificial Analysis chart: DeepSeek V4 Pro output speed vs price, Systalyze first at 277 t/s and $0.33
DeepSeek V4 Pro · Output speed vs price. Screenshot from Artificial Analysis.
End-to-end response time vs price

Fastest and cheapest of 11 providers.

Lowest response time and lowest price of 11 providers. 1.4× faster than Nebius at 4.7× lower price, and 3.6× faster than DeepSeek's own API at 2.0× lower price.

End-to-end response time vs price. Seconds to receive a 500-token answer plotted against blended price per 1M tokens. Lower and further left is better.

Artificial Analysis chart: DeepSeek V4 Pro end-to-end response time vs price, Systalyze first at 10.4 s and $0.33
DeepSeek V4 Pro · End-to-end response time vs price. Screenshot from Artificial Analysis.

DeepSeek V4 Flash

Output speed

332 tokens per second

#1 of 16 providers on output speed. 1.5× faster than DeepSeek's own API and 6.5× faster than Nebius.

Best gain 6.5×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: DeepSeek V4 Flash output speed, Systalyze first at 332 tokens per second
DeepSeek V4 Flash · Output speed. Screenshot from Artificial Analysis.

Muse Glimmer

Output speed

161 tokens per second

#1 of 4 providers on output speed: 1.5× Together AI and 2.5× Fireworks.

Best gain 2.5×

Output speed. How many tokens per second the model produces once it starts writing, measured with a 10,000-token prompt. This is raw generation speed. Higher is better.

Artificial Analysis chart: Muse Glimmer, Systalyze first at 161 tokens per second
Muse Glimmer · Output speed. Screenshot from Artificial Analysis.
Blended price

$0.13 per 1M tokens, blended

#1 of 4 providers on blended price. 1.9× cheaper than Together AI and Fireworks, and 1.6× cheaper than the next cheapest provider.

Best gain 1.9×

Blended price. The price per 1M tokens, blending cache hit, input and output prices at a 7:2:1 ratio. Lower is better.

Artificial Analysis chart: Muse Glimmer blended price, Systalyze first at $0.13
Muse Glimmer · Blended price. Screenshot from Artificial Analysis.
Token prices

$1.00 per 1M output tokens

#1 of 4 providers on output price. 1.5× cheaper than Together AI and Fireworks on output, with input at $0.10 (3.5× cheaper than their $0.35) and cache hits at $0.01 (4.0× cheaper than their $0.04).

Best gain 1.5×

Token prices. The list price per 1M tokens for cache hits, input and output, measured with 10,000 input tokens. The table shows output price. Lower is better.

Artificial Analysis chart: Muse Glimmer token prices, Systalyze first at $1.00 per 1M output tokens
Muse Glimmer · Token prices. Screenshot from Artificial Analysis.
Time to first answer token

14.0 seconds to first answer token

#1 of 4 providers on time to first answer token. 1.4× faster than Together AI and 2.3× faster than Fireworks.

Best gain 2.3×

Time to first answer token. How many seconds pass before the first answer token arrives, including input processing and any thinking time, measured with a 10,000-token prompt. Lower is better.

Artificial Analysis chart: Muse Glimmer time to first answer token, Systalyze first at 14.0 s
Muse Glimmer · Time to first answer token. Screenshot from Artificial Analysis.
End-to-end response time

17.1 seconds end to end

#1 of 4 providers on end-to-end response time. 1.4× faster than Together AI and 2.3× faster than Fireworks.

Best gain 2.3×

End-to-end response time. How many seconds it takes to return 500 output tokens, including input processing and thinking time, measured with a 10,000-token prompt. Lower is better.

Artificial Analysis chart: Muse Glimmer end-to-end response time, Systalyze first at 17.1 s
Muse Glimmer · End-to-end response time. Screenshot from Artificial Analysis.
Response time vs price

Fastest and cheapest of 4 providers.

#1 of 4 providers on both end-to-end response time and blended price. Together AI and Fireworks cost about twice as much and take 25 to 41 seconds.

End-to-end response time vs price. Each point is a provider: seconds to return 500 output tokens against blended price per 1M tokens. Lower and further left is better.

Artificial Analysis chart: Muse Glimmer response time vs price, Systalyze first at about 17 s and $0.13
Muse Glimmer · Response time vs price. Screenshot from Artificial Analysis.
Output speed vs price

Fastest and cheapest of 4 providers.

#1 of 4 providers on both output speed and blended price. About 1.5× the output speed of Together AI and 2.5× that of Fireworks, at roughly half their price.

Output speed vs price. Each point is a provider: output tokens per second against blended price per 1M tokens, blended 7:2:1 cache hit to input to output. Higher and further left is better.

Artificial Analysis chart: Muse Glimmer output speed vs price, Systalyze first at about 161 t/s and $0.13
Muse Glimmer · Output speed vs price. Screenshot from Artificial Analysis.

About these numbers.

Every figure on this page is measured by Artificial Analysis, an independent benchmarking firm. They run the same prompts against each provider’s public endpoint and publish the results on their site.

Gains are against the labs’ own APIs and the major inference providers.

Run your models this fast.