Model performance on MindsHub

See how models served through MindsHub Inference perform: response times, availability and cost, with current prices, API aliases and fallback routes.

Measurements refresh every 2 to 6 hours · About the data

All makers
Makers
Reasoning effort

Leaderboard

Compare typical results. Expand a model for more details and data at every effort level.

30 days7 days24 hours

Showing 46 models

What the columns mean
Model

Expand a model for more details and data for every reasoning effort level.

About model aliases

Intelligence

MindsHub’s read is a subjective estimate from 1 to 10 based on public benchmarks and rankings, not a precise measurement. Performance depends on your workload; test models on your own tasks.

For detailed benchmarks, refer to third party services such as Artificial Analysis, Epoch AI or similar.

About the intelligence estimate

Ping · p50

Time to the first answer token for a one-word reply at the lowest reasoning effort. This checks basic responsiveness.

Includes thinking before the answer starts, not the time to finish it. Lower is faster. P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements.

Ping methodology and sample prompt

Simple task · p50

Time to the first answer token at the default reasoning effort. The model interprets a short briefing and reasons about how to describe it in at least 150 words, using only the supplied facts.

Includes thinking before the answer starts, not the time to finish it. Lower is faster. P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements.

Simple task methodology and sample prompt

Complex task · p50

Time to the first answer token at the selected reasoning effort. The model analyzes a longer briefing and plans an 800+ word guide that covers its facts in order and follows content and formatting rules.

Includes thinking before the answer starts, not the time to finish it. Lower is faster. P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements.

Complex task methodology and sample prompt

Writing speed

How fast the model writes after its answer starts, in tokens per second. Measured on the 800+ word task. Higher is faster; thinking time is excluded.

How writing speed is measured

Context limit

How much input the model can take in one request, measured in tokens. Tokens are pieces of text.

About tokens, context and prices

Cost per task

Cost of the same writing task relative to the most expensive measured model and effort level (100%). 50% costs half as much. The collapsed range covers measured effort levels; expand for each level. Lower is cheaper.

100% = Claude Opus 5.5 at max, the highest measured cost for this writing task.

How cost per task is compared

Token price

US dollars per million tokens: input / output. Input is what you send; output is what the model generates. See Prices for other rates.

About token prices and the free tier

Availability

Below 100% does not necessarily mean customer requests failed: a configured fallback route can still answer. Share of automated checks answered by the model you chose, including retries. Other checks were answered by a fallback or got no answer. Failures attributed to our monitoring network are excluded. Higher is better.

How availability and fallbacks work

First answer token is the wait before the answer starts, including thinking. P50 is the typical result. Cost per task spans the measured effort levels.

Every model MindsHub serves: intelligence (our estimate, from 1 to 10), time to first word at three tasks, writing speed, context limit, relative cost per task, token price and availability before fallback. Sorted by intelligence.

Model

Expand a model for more details and data for every reasoning effort level.

About model aliases

Intelligence

MindsHub’s read is a subjective estimate from 1 to 10 based on public benchmarks and rankings, not a precise measurement. Performance depends on your workload; test models on your own tasks.

For detailed benchmarks, refer to third party services such as Artificial Analysis, Epoch AI or similar.

About the intelligence estimate

Ping · p50

Time to the first answer token for a one-word reply at the lowest reasoning effort. This checks basic responsiveness.

Includes thinking before the answer starts, not the time to finish it. Lower is faster. P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements.

Ping methodology and sample prompt

Simple task · p50

Time to the first answer token at the default reasoning effort. The model interprets a short briefing and reasons about how to describe it in at least 150 words, using only the supplied facts.

Includes thinking before the answer starts, not the time to finish it. Lower is faster. P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements.

Simple task methodology and sample prompt

Complex task · p50

Time to the first answer token at the selected reasoning effort. The model analyzes a longer briefing and plans an 800+ word guide that covers its facts in order and follows content and formatting rules.

Includes thinking before the answer starts, not the time to finish it. Lower is faster. P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements.

Complex task methodology and sample prompt

Writing speed

How fast the model writes after its answer starts, in tokens per second. Measured on the 800+ word task. Higher is faster; thinking time is excluded.

How writing speed is measured

Context limit

How much input the model can take in one request, measured in tokens. Tokens are pieces of text.

About tokens, context and prices

Cost per task

Cost of the same writing task relative to the most expensive measured model and effort level (100%). 50% costs half as much. The collapsed range covers measured effort levels; expand for each level. Lower is cheaper.

100% = Claude Opus 5.5 at max, the highest measured cost for this writing task.

How cost per task is compared

Token price

US dollars per million tokens: input / output. Input is what you send; output is what the model generates. See Prices for other rates.

About token prices and the free tier

Availability

Below 100% does not necessarily mean customer requests failed: a configured fallback route can still answer. Share of automated checks answered by the model you chose, including retries. Other checks were answered by a fallback or got no answer. Failures attributed to our monitoring network are excluded. Higher is better.

How availability and fallbacks work

Older version
Anthropic
8.5 9.5 10 1.33 s 3.42 s — 5.37 s — — 84 — 1M 10% $8 / $40 100.00% 100.00% 100.00%
Claude Fable 5.2, effortlow 8.5 1.33 s — — —
Claude Fable 5.2, effortmedium 9 — — —
Claude Fable 5.2, efforthigh default 9.5 3.42 s 5.37 s 84 10.2%
Claude Fable 5.2, effortxhigh 9.5 — — —
Claude Fable 5.2, effortmax 10 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
fable-5-2
Other versions
fable-5-1 fable-5
Also accepts
claude-fable-5-2 fable-5.2
Fallbacks
1st fallback: GPT
Inputs
Text
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 10–18
Closed
Anthropic
7 8.5 9.5 1.44 s 8.11 s 1.74 s 6.99 s 313 s 81 80 110 1M 5%–100% $4 / $20 98.37% 100.00% 100.00%
Claude Opus 5.5, effortlow 7 1.44 s 1.74 s 81 5.1%
Claude Opus 5.5, effortmedium 8 3.64 s 79 5.6%
Claude Opus 5.5, efforthigh default 8.5 8.11 s 6.99 s 80 6.7%
Claude Opus 5.5, effortxhigh 9 40.3 s 96 14.8%
Claude Opus 5.5, effortmax 9.5 313 s 110 100.0%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
opus-5-5
Rotating
opus
Other versions
opus-5
Also accepts
opus-5.5
Fallbacks
1st fallback: GPT
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
OpenAI
7.5 8.5 9.5 1.98 s 4.68 s 2.01 s 5.78 s 90.7 s 54 38 71 1.05M 9%–31% $10 / $50 100.00% 100.00% 100.00%
GPT-6 Astra, effortlow 7.5 1.98 s 2.01 s 54 8.7%
GPT-6 Astra, effortmedium default 8.5 4.68 s 5.78 s 38 9.1%
GPT-6 Astra, efforthigh 9 15.3 s 56 11.3%
GPT-6 Astra, effortxhigh 9 41.0 s 51 15.4%
GPT-6 Astra, effortmax 9.5 90.7 s 71 30.9%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-6-astra
Rotating
gpt
Other versions
gpt-5-6-sol
Fallbacks
1st fallback: Claude Opus2nd fallback: Gemini Pro
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
Anthropic
7 8.5 9 1.23 s 4.33 s 1.54 s 3.03 s 92.5 s 61 58 83 1M 12%–70% $10 / $50 98.38% 100.00% 100.00%
Claude Fable 5.1, effortlow 7 1.23 s 1.54 s 61 11.8%
Claude Fable 5.1, effortmedium 8 2.14 s 58 12.9%
Claude Fable 5.1, efforthigh default 8.5 4.33 s 3.03 s 58 14.4%
Claude Fable 5.1, effortxhigh 9 16.3 s 59 25.9%
Claude Fable 5.1, effortmax 9 92.5 s 83 69.6%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
fable-5-1
Rotating
fable
Other versions
fable-5-2 fable-5
Also accepts
fable-5.1
Fallbacks
1st fallback: GPT
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
Anthropic
5 7 9 1.56 s 4.00 s 1.60 s 4.45 s 173 s 130 129 160 1M 2%–38% $2 / $10 98.38% 100.00% 100.00%
Claude Sonnet 5.5, effortlow 5 1.56 s 1.60 s 130 2.1%
Claude Sonnet 5.5, effortmedium 6 1.61 s 125 2.1%
Claude Sonnet 5.5, efforthigh default 7 4.00 s 4.45 s 129 2.5%
Claude Sonnet 5.5, effortxhigh 8 7.78 s 127 3.3%
Claude Sonnet 5.5, effortmax 9 173 s 160 37.9%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
sonnet-5-5
Rotating
sonnet
Other versions
sonnet-5
Also accepts
sonnet-5.5
Fallbacks
1st fallback: GPT
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
Google
7.5 9 5.70 s 5.27 s 4.79 s 79.6 s 73 90 1.05M 2%–10% $2 / $10 99.88% 99.92% 99.53%
Gemini 4 Argon, effortlow default 7.5 5.70 s 5.27 s 4.79 s 73 2.3%
Gemini 4 Argon, effortmedium 8 4.35 s 73 2.3%
Gemini 4 Argon, efforthigh 9 79.6 s 90 10.5%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-argon-4
Rotating
gemini-argon
Fallbacks
1st fallback: Claude Sonnet2nd fallback: GPT
Inputs
Text, images, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Oct 17–Nov 18
Licence
Closed.
Closed
OpenAI
6 7.5 9 1.98 s 3.42 s 1.57 s 2.98 s 68.6 s 66 61 77 1.05M 2%–6% $2 / $10 99.99% 100.00% 100.00%
GPT-6.1 Sol, effortlow 6 1.98 s 1.57 s 66 1.7%
GPT-6.1 Sol, effortmedium default 7.5 3.42 s 2.98 s 61 1.8%
GPT-6.1 Sol, efforthigh 8 10.5 s 68 2.1%
GPT-6.1 Sol, effortxhigh 8.5 23.6 s 64 2.9%
GPT-6.1 Sol, effortmax 9 68.6 s 77 5.6%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-sol-6-1
Rotating
gpt-sol
Other versions
gpt-sol-6
Fallbacks
1st fallback: Claude Opus2nd fallback: Gemini Pro
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Nov 4–18
Licence
Closed.
Closed Older version
Anthropic
6.5 8 8.5 2.71 s 3.16 s — 3.13 s — — 60 — 1M 12% $10 / $50 98.37% 100.00% 100.00%
Claude Fable 5, effortlow 6.5 2.71 s — — —
Claude Fable 5, effortmedium 7.5 — — —
Claude Fable 5, efforthigh default 8 3.16 s 3.13 s 60 12.1%
Claude Fable 5, effortxhigh 8.5 — — —
Claude Fable 5, effortmax 8.5 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
fable-5
Other versions
fable-5-1 fable-5-2
Also accepts
claude-fable-5
Fallbacks
1st fallback: GPT
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 21–Nov 18
Licence
Closed.
Closed Older version
Anthropic
6.5 7.5 8 1.62 s — — 3.33 s — — 56 — 1M 6% $5 / $25 98.36% 100.00% 100.00%
Claude Opus 5, effortlow 6.5 1.62 s — — —
Claude Opus 5, effortmedium 7 — — —
Claude Opus 5, efforthigh default 7.5 3.33 s 56 6.4%
Claude Opus 5, effortxhigh 8 — — —
Claude Opus 5, effortmax 8 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
opus-5
Other versions
opus-5-5
Also accepts
claude-opus-5
Fallbacks
1st fallback: GPT
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsComplex task: 40 test runs, Oct 16–Nov 18
Licence
Closed.
Closed Older version
OpenAI
4.5 6.5 8 1.50 s 3.04 s — 2.97 s — — 126 — 1.05M 5% $5 / $30 99.98% 100.00% 100.00%
GPT 5.6 Sol, effortnone 4.5 1.50 s — — —
GPT 5.6 Sol, effortlow 5.5 — — —
GPT 5.6 Sol, effortmedium default 6.5 3.04 s 2.97 s 126 5.3%
GPT 5.6 Sol, efforthigh 7.5 — — —
GPT 5.6 Sol, effortxhigh 8 — — —
GPT 5.6 Sol, effortmax 8 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-5-6-sol
Other versions
gpt-6-astra
Also accepts
gpt-5.6-sol
Fallbacks
1st fallback: Claude Opus2nd fallback: Gemini Pro
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 4–17
Licence
Closed.
Closed Older version
OpenAI
4.5 6.5 8 1.39 s 3.85 s — 2.23 s — — 130 — 1.05M 2% $2 / $10 100.00% 100.00% 100.00%
GPT-6 Sol, effortnone 4.5 1.39 s — — —
GPT-6 Sol, effortlow 5.5 — — —
GPT-6 Sol, effortmedium default 6.5 3.85 s 2.23 s 130 1.8%
GPT-6 Sol, efforthigh 7.5 — — —
GPT-6 Sol, effortxhigh 8 — — —
GPT-6 Sol, effortmax 8 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-sol-6
Other versions
gpt-sol-6-1
Also accepts
gpt-6-sol
Fallbacks
1st fallback: Claude Opus2nd fallback: Gemini Pro
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 4–17
Licence
Closed.
Closed
OpenAI
4 5.5 7.5 1.69 s 6.41 s 1.53 s 3.10 s 157 s 89 90 87 1.05M <1%–9% $1 / $5 100.00% 100.00% 100.00%
GPT-6.1 Terra, effortnone 4 1.69 s 1.53 s 89 0.7%
GPT-6.1 Terra, effortlow 4.5 1.90 s 90 0.7%
GPT-6.1 Terra, effortmedium default 5.5 6.41 s 3.10 s 90 0.8%
GPT-6.1 Terra, efforthigh 6 7.31 s 97 1.0%
GPT-6.1 Terra, effortxhigh 6.5 28.9 s 95 2.1%
GPT-6.1 Terra, effortmax 7.5 157 s 87 8.8%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-terra-6-1
Rotating
gpt-terra
Other versions
gpt-terra-5-6
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Gemini Pro
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 240 test runs, Oct 29–Nov 18
Licence
Closed.
Closed
SpaceXAI
6 7 7.5 1.90 s 40.5 s 5.52 s 38.7 s 135 s 81 85 117 500K 2%–13% $2 / $6 99.93% 100.00% 100.00%
Grok 4.7, effortlow 6 1.90 s 5.52 s 81 1.9%
Grok 4.7, effortmedium 6.5 10.0 s 80 2.7%
Grok 4.7, efforthigh default 7 40.5 s 38.7 s 85 6.1%
Grok 4.7, effortxhigh 7.5 135 s 117 12.9%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
grok-4-7
Rotating
grok
Other versions
grok-4-6 grok-4-5
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 160 test runs, Oct 16–Nov 18
Licence
Closed.
API only
Moonshot AI
7.5 3.42 s 43.8 s 53.7 s 38 1M 4% $2 / $10 98.33% 98.55% 96.95%
Kimi K3.1 Pro, no effort setting 7.5 3.42 s 43.8 s 53.7 s 38 3.6%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
kimi-k3-1-pro
Rotating
kimi-pro
Fallbacks
1st fallback: DeepSeek Flash
Inputs
Text, images, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 27–Nov 17
Licence
API only.
Open weights
Moonshot AI · via Fireworks AI
7 2.40 s 13.0 s 21.5 s 59 1M 5% $3 / $15 99.87% 100.00% 100.00%
Kimi K3, no effort setting 7 2.40 s 13.0 s 21.5 s 59 5.3%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
kimi-k3
Rotating
kimi
Fallbacks
1st fallback: DeepSeek Flash
Inputs
Text, images
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 4–17
Licence
Open weights · Kimi K3 License. The licence is Moonshot's own, not MIT or Apache. Commercial use is allowed with conditions that start at scale, including attribution for very large products and a separate agreement for large model-as-a-service operators, so read it before you build on it.
Closed
Meta
7 2.22 s 26.3 s 59.5 s 431 1.05M 28% $4 / $20 99.99% 100.00% 100.00%
Muse Wonder 1.0, no effort setting 7 2.22 s 26.3 s 59.5 s 431 28.5%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
muse-wonder-1-0
Rotating
muse-wonder
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
Anthropic
4 5 6.5 1.14 s 4.00 s 2.36 s 2.87 s 136 s 185 194 263 1M <1%–2% $0.10 / $0.50 99.98% 100.00% 100.00%
Claude Haiku 5.5, effortlow 4 1.14 s 2.36 s 185 0.1%
Claude Haiku 5.5, effortmedium default 5 4.00 s 2.87 s 194 0.1%
Claude Haiku 5.5, efforthigh 5.5 5.85 s 191 0.2%
Claude Haiku 5.5, effortxhigh 6 13.6 s 221 0.3%
Claude Haiku 5.5, effortmax 6.5 136 s 263 2.4%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
haiku-5-5
Rotating
haiku
Other versions
haiku-4-5
Also accepts
haiku-5.5
Fallbacks
1st fallback: Gemini Flash2nd fallback: DeepSeek Flash
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Nov 7–18
Licence
Closed.
Open weights
DeepSeek · via Fireworks AI
4.5 6 6.5 0.98 s 4.87 s 1.10 s 11.4 s 23.1 s 304 290 299 1M <1% $0.30 / $1.20 99.96% 99.96% 100.00%
DeepSeek V4.1 Flash 1017, effortnone 4.5 0.98 s 1.10 s 304 0.2%
DeepSeek V4.1 Flash 1017, effortlow 5 9.74 s 289 0.5%
DeepSeek V4.1 Flash 1017, efforthigh default 6 4.87 s 11.4 s 290 0.6%
DeepSeek V4.1 Flash 1017, effortmax 6.5 23.1 s 299 1.0%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
deepseek-v4-1-flash-1017
Rotating
deepseek-v4-flash
Other versions
deepseek-v4-1-flash
Fallbacks
1st fallback: GLM Flash
Inputs
Text, images
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 160 test runs, Oct 27–Nov 18
Licence
Open weights · MIT.
Closed Older version
Google
5 5.5 6.5 2.22 s 25.3 s — 33.9 s — — 138 — 1.05M 3% $0.75 / $3.75 99.99% 100.00% 100.00%
Gemini 3.7 Flash, effortlow 5 2.22 s — — —
Gemini 3.7 Flash, effortmedium default 5.5 25.3 s 33.9 s 138 2.6%
Gemini 3.7 Flash, efforthigh 6.5 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-flash-3-7
Also accepts
gemini-3.7-flash gemini-flash-3.7
Fallbacks
1st fallback: GPT Mini2nd fallback: DeepSeek Flash
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 17
Licence
Closed.
Closed
Google
5 6 6.5 2.31 s 20.0 s 2.50 s 34.6 s 59.7 s 120 135 168 1.05M <1%–4% $0.75 / $3.75 99.99% 100.00% 100.00%
Gemini 3.8 Flash, effortlow 5 2.31 s 2.50 s 120 0.9%
Gemini 3.8 Flash, effortmedium default 6 20.0 s 34.6 s 135 2.6%
Gemini 3.8 Flash, efforthigh 6.5 59.7 s 168 4.3%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-flash-3-8
Rotating
gemini-flash
Also accepts
gemini-flash-3.8
Fallbacks
1st fallback: GPT Mini2nd fallback: DeepSeek Flash
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Oct 16–Nov 18
Licence
Closed.
Open weights
Zhipu · via Fireworks AI
5 6.5 1.01 s 28.5 s 1.30 s 68.9 s 122 146 1M <1%–5% $1.40 / $4.40 99.32% 100.00% 100.00%
GLM 5.3, effortlow 5 1.01 s 1.30 s 122 0.8%
GLM 5.3, efforthigh 6 1.45 s 122 0.8%
GLM 5.3, effortmax default 6.5 28.5 s 68.9 s 146 5.2%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
glm-5-3
Rotating
glm
Also accepts
glm-5.3
Fallbacks
1st fallback: DeepSeek Flash
Inputs
Text
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Nov 4–18
Licence
Open weights · GLM-5.3 License. Zhipu released the weights under its own GLM-5.3 License, which allows commercial use with conditions: read the terms before you build on it.
Closed Older version
OpenAI
3.5 5 6.5 1.33 s 1.62 s — 1.74 s — — 111 — 1.05M 2% $2 / $12 99.98% 100.00% 100.00%
GPT 5.6 Terra, effortnone 3.5 1.33 s — — —
GPT 5.6 Terra, effortlow 4 — — —
GPT 5.6 Terra, effortmedium default 5 1.62 s 1.74 s 111 2.0%
GPT 5.6 Terra, efforthigh 6 — — —
GPT 5.6 Terra, effortxhigh 6.5 — — —
GPT 5.6 Terra, effortmax 6.5 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-terra-5-6
Other versions
gpt-terra-6-1
Also accepts
gpt-terra-5.6 gpt-5.6-terra
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Gemini Pro
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 17
Licence
Closed.
Closed Older version
SpaceXAI
5.5 6.5 6.5 2.71 s 53.3 s — 75.6 s — — 104 — 500K 6% $2 / $6 99.89% 100.00% 100.00%
Grok 4.6, effortlow 5.5 2.71 s — — —
Grok 4.6, effortmedium 6 — — —
Grok 4.6, efforthigh default 6.5 53.3 s 75.6 s 104 5.9%
Grok 4.6, effortxhigh 6.5 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
grok-4-6
Other versions
grok-4-7 grok-4-5
Also accepts
grok-4.6
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 4–17
Licence
Closed.
Closed
Meta
6.5 3.85 s 55.4 s 55.8 s 141 1.05M 5% $1.25 / $4.25 99.98% 100.00% 100.00%
Muse Spark 1.3, no effort setting 6.5 3.85 s 55.4 s 55.8 s 141 4.7%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
muse-spark-1-3
Rotating
muse-spark
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 11–18
Licence
Closed.
Closed Older version
Anthropic
4 5 6 1.19 s 1.33 s — 1.34 s — — 69 — 1M 2% $2 / $10 98.39% 100.00% 100.00%
Claude Sonnet 5, effortlow 4 1.19 s — — —
Claude Sonnet 5, effortmedium 4.5 — — —
Claude Sonnet 5, efforthigh default 5 1.33 s 1.34 s 69 2.4%
Claude Sonnet 5, effortmax 6 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
sonnet-5
Other versions
sonnet-5-5
Also accepts
claude-sonnet-5
Fallbacks
1st fallback: GPT
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 11–18
Licence
Closed.
Open weights Older version
DeepSeek · via Fireworks AI
4 5.5 6 0.77 s 11.1 s — 32.9 s — — 270 — 1M 1% $0.30 / $1.20 99.79% 99.68% 100.00%
DeepSeek V4.1 Flash, effortnone 4 0.77 s — — —
DeepSeek V4.1 Flash, effortlow 4.5 — — —
DeepSeek V4.1 Flash, efforthigh default 5.5 11.1 s 32.9 s 270 1.1%
DeepSeek V4.1 Flash, effortmax 6 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
deepseek-v4-1-flash
Other versions
deepseek-v4-1-flash-1017
Also accepts
accounts/fireworks/models/deepseek-v4p1-flash deepseek-v4p1-flash
Fallbacks
1st fallback: GLM Flash
Inputs
Text, images
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 11–18
Licence
Open weights · MIT.
Closed Older version
Google
4 5.5 6 1.39 s 16.4 s — 27.6 s — — 229 — 1.05M 7% $1.50 / $9 99.99% 100.00% 100.00%
Gemini 3.5 Flash, effortminimal 4 1.39 s — — —
Gemini 3.5 Flash, effortlow 5 — — —
Gemini 3.5 Flash, effortmedium default 5.5 16.4 s 27.6 s 229 7.0%
Gemini 3.5 Flash, efforthigh 6 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-flash-3-5
Also accepts
gemini-3.5-flash gemini-flash-3.5
Fallbacks
1st fallback: GPT Mini2nd fallback: DeepSeek Flash
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 17
Licence
Closed.
Open weights
Zhipu · via Fireworks AI
4.5 6 1.28 s 14.0 s 1.17 s 37.7 s 190 202 1M <1% $0.15 / $0.50 99.91% 100.00% 100.00%
GLM 5.3 Flash, effortlow 4.5 1.28 s 1.17 s 190 <0.1%
GLM 5.3 Flash, efforthigh 5.5 1.43 s 171 <0.1%
GLM 5.3 Flash, effortmax default 6 14.0 s 37.7 s 202 0.5%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
glm-flash-5-3
Rotating
glm-5-3-flash
Also accepts
glm-flash-5.3
Fallbacks
1st fallback: DeepSeek Flash
Inputs
Text, images
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Oct 16–Nov 18
Licence
Open weights · MIT.
Closed Older version
SpaceXAI
5 6 1.19 s 11.5 s — 8.16 s — 69 500K 2% $2 / $6 99.84% 99.96% 100.00%
Grok 4.5, effortlow 5 1.19 s — — —
Grok 4.5, effortmedium 5.5 — — —
Grok 4.5, efforthigh default 6 11.5 s 8.16 s 69 2.0%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
grok-4-5
Other versions
grok-4-7 grok-4-6
Also accepts
grok-4.5
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 4–17
Licence
Closed.
Meta
6 1.39 s 6.41 s 12.1 s 1,199 1.05M <1% $0.10 / $0.50 100.00% 100.00% 100.00%
Muse Ember 1.0, no effort setting 6 1.39 s 6.41 s 12.1 s 1,199 0.5%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
muse-ember-1-0
Rotating
muse-ember
Fallbacks
1st fallback: Gemini Flash2nd fallback: DeepSeek Flash
Inputs
Text
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 10–18
Closed Older version
Meta
6 3.16 s 21.6 s 22.0 s 278 1.05M 3% $1.25 / $4.25 100.00% 100.00% 100.00%
Muse Spark 1.2, no effort setting 6 3.16 s 21.6 s 22.0 s 278 3.0%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
muse-spark-1-2
Also accepts
muse-spark-1.2
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 11–18
Licence
Closed.
Closed Older version
Google
4 5 5.5 1.39 s 13.5 s — 26.4 s — — 229 — 1.05M 3% $0.75 / $3.75 100.00% 100.00% 100.00%
Gemini 3.6 Flash, effortminimal 4 1.39 s — — —
Gemini 3.6 Flash, effortlow 4.5 — — —
Gemini 3.6 Flash, effortmedium default 5 13.5 s 26.4 s 229 2.9%
Gemini 3.6 Flash, efforthigh 5.5 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-flash-3-6
Also accepts
gemini-3.6-flash gemini-flash-3.6
Fallbacks
1st fallback: GPT Mini2nd fallback: DeepSeek Flash
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 17
Licence
Closed.
Closed
OpenAI
4 4.5 5.5 1.44 s 2.71 s 1.83 s 3.61 s 22.0 s 76 80 84 400K 3%–6% $1.75 / $14 99.99% 100.00% 100.00%
GPT 5.3 Codex, effortlow 4 1.44 s 1.83 s 76 2.7%
GPT 5.3 Codex, effortmedium default 4.5 2.71 s 3.61 s 80 2.9%
GPT 5.3 Codex, efforthigh 5 10.8 s 82 3.9%
GPT 5.3 Codex, effortxhigh 5.5 22.0 s 84 6.3%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-codex-5-3
Rotating
gpt-codex
Also accepts
gpt-codex-5.3
Fallbacks
1st fallback: Claude Sonnet
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 160 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
OpenAI
2.5 4 5.5 1.56 s 2.71 s 1.30 s 3.47 s 15.7 s 115 131 128 1.05M <1% $0.10 / $0.50 99.99% 100.00% 100.00%
GPT-6 Luna, effortnone 2.5 1.56 s 1.30 s 115 <0.1%
GPT-6 Luna, effortlow 3 2.23 s 129 <0.1%
GPT-6 Luna, effortmedium default 4 2.71 s 3.47 s 131 <0.1%
GPT-6 Luna, efforthigh 4.5 4.62 s 129 <0.1%
GPT-6 Luna, effortxhigh 5 6.09 s 128 <0.1%
GPT-6 Luna, effortmax 5.5 15.7 s 128 0.2%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-luna-6
Rotating
gpt-luna
Other versions
gpt-luna-5-6
Fallbacks
1st fallback: Gemini Flash2nd fallback: Claude Haiku
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 240 test runs, Oct 16–Nov 18
Licence
Closed.
Closed Older version
Meta
5.5 3.16 s 13.5 s 15.3 s 566 1.05M 3% $1.25 / $4.25 99.99% 100.00% 100.00%
Muse Spark 1.1, no effort setting 5.5 3.16 s 13.5 s 15.3 s 566 3.3%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
muse-spark-1-1
Also accepts
muse-spark-1.1
Fallbacks
1st fallback: Claude Sonnet2nd fallback: Kimi
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 21–Nov 18
Licence
Closed.
Open weights
Alibaba · via Fireworks AI
4.5 5.5 1.44 s 22.5 s 2.55 s 71.2 s 271 234 262K 1%–12% $2 / $6 98.07% 99.71% 100.00%
Qwen3.8-2.4T-A95B, effortlow 4.5 1.44 s 2.55 s 271 1.2%
Qwen3.8-2.4T-A95B, effortmedium 5 4.29 s 231 1.6%
Qwen3.8-2.4T-A95B, effortxhigh default 5.5 22.5 s 71.2 s 234 11.5%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
qwen-3-8-a95b
Rotating
qwen
Fallbacks
1st fallback: DeepSeek Flash
Inputs
Text
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Oct 21–Nov 18
Licence
Open weights · Qwen3.8-Max License. The Qwen3.8-Max License allows commercial use with conditions, including separate authorization for large model-as-a-service and AI work-assistant businesses, so read it before you build on it.
Closed
Google
4 5 3.70 s 29.6 s 29.3 s 44.7 s 211 199 1.05M 7%–12% $2 / $12 99.94% 100.00% 100.00%
Gemini 3.1 Pro Preview, effortlow 4 3.70 s 29.3 s 211 7.0%
Gemini 3.1 Pro Preview, effortmedium 4.5 33.8 s 215 9.0%
Gemini 3.1 Pro Preview, efforthigh default 5 29.6 s 44.7 s 199 12.0%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-3-1-pro
Rotating
gemini
Fallbacks
1st fallback: Claude Sonnet2nd fallback: GPT
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Oct 29–Nov 18
Licence
Closed.
Closed Older version
OpenAI
2.5 4 5 1.28 s 1.83 s — 1.91 s — — 121 — 1.05M <1% $0.20 / $1.20 100.00% 100.00% 100.00%
GPT 5.6 Luna, effortnone 2.5 1.28 s — — —
GPT 5.6 Luna, effortlow 3.5 — — —
GPT 5.6 Luna, effortmedium default 4 1.83 s 1.91 s 121 0.2%
GPT 5.6 Luna, efforthigh 4.5 — — —
GPT 5.6 Luna, effortxhigh 5 — — —
GPT 5.6 Luna, effortmax 5 — — —
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-luna-5-6
Other versions
gpt-luna-6
Also accepts
gpt-5.6-luna gpt-luna-5.6
Fallbacks
1st fallback: Gemini Flash2nd fallback: Claude Haiku
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 17
Licence
Closed.
Closed Older version
Google
2.5 4 1.14 s 10.3 s — 16.1 s — 217 1.05M 2% $0.50 / $3 100.00% 100.00% 100.00%
Gemini 3 Flash Preview, effortminimal 2.5 1.14 s — — —
Gemini 3 Flash Preview, effortlow 3 — — —
Gemini 3 Flash Preview, effortmedium 3.5 — — —
Gemini 3 Flash Preview, efforthigh default 4 10.3 s 16.1 s 217 1.8%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-flash-3
Also accepts
gemini-3-flash-preview
Fallbacks
1st fallback: GPT Mini2nd fallback: DeepSeek Flash
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 15–Nov 17
Licence
Closed.
Closed
OpenAI
2 4 1.28 s 1.14 s 1.13 s 69.5 s 218 260 400K <1%–10% $0.75 / $4.50 99.99% 100.00% 100.00%
GPT 5.4 Mini, effortnone default 2 1.28 s 1.14 s 1.13 s 218 0.8%
GPT 5.4 Mini, effortlow 3 1.36 s 220 0.8%
GPT 5.4 Mini, effortmedium 3.5 2.34 s 214 0.9%
GPT 5.4 Mini, efforthigh 4 6.69 s 218 1.4%
GPT 5.4 Mini, effortxhigh 4 69.5 s 260 9.9%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-mini-5-4
Rotating
gpt-mini
Also accepts
gpt-mini-5.4
Fallbacks
1st fallback: Gemini Flash2nd fallback: Claude Haiku
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Oct 16–Nov 18
Licence
Closed.
Closed Free tier
MindsHub
2.5 4 1.56 s 2.93 s 1.42 s 3.40 s 112 126 1.05M <1% $0.10 / $0.50 100.00% 100.00% 100.00%
MindsHub Air, effortnone 2.5 1.56 s 1.42 s 112 <0.1%
MindsHub Air, effortlow 3 2.22 s 124 <0.1%
MindsHub Air, effortmedium default 4 2.93 s 3.40 s 126 <0.1%
Response times
Typical time to the first answer token (p50), including thinking.
What MindsHub Air is
Our rotating alias for everyday knowledge work. The underlying model may change as better options become available.
Rotating alias
mindshub_air
Fallbacks
1st fallback: Gemini Flash
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 120 test runs, Oct 20–Nov 18
Licence
Closed.
API only
MindsHub · via Cerebras
2.5 4 0.59 s — 0.76 s 8.61 s 1,381 2,967 131K <1%–3% $0.99 / $1.49 99.93% 99.96% 100.00%
MindsHub Blaze, effortnone 2.5 0.59 s 0.76 s 1,381 0.4%
MindsHub Blaze, efforthigh default 4 8.61 s 2,967 3.3%
Response times
Typical time to the first answer token (p50), including thinking.
What MindsHub Blaze is
Our rotating alias for speed, served by Cerebras. We keep it pointed at a model optimized for speed, so the model behind it can change.
Rotating alias
mindshub_blaze
Fallbacks
1st fallback: GPT-OSS 120B (Fireworks)
Inputs
Text, images
Served by
Cerebras
Measured
Ping: every 10 minutes, from multiple regionsComplex task: 80 test runs, Oct 29–Nov 18
Licence
API only.
Closed
OpenAI
1.5 3 1.19 s 0.98 s 1.06 s 25.0 s 165 171 400K <1% $0.20 / $1.25 99.93% 100.00% 100.00%
GPT 5.4 Nano, effortnone default 1.5 1.19 s 0.98 s 1.06 s 165 0.3%
GPT 5.4 Nano, effortlow 2 1.21 s 167 0.3%
GPT 5.4 Nano, effortmedium 2.5 3.78 s 166 0.3%
GPT 5.4 Nano, efforthigh 3 4.11 s 167 0.3%
GPT 5.4 Nano, effortxhigh 3 25.0 s 171 0.9%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-nano-5-4
Rotating
gpt-nano
Also accepts
gpt-nano-5.4
Fallbacks
1st fallback: Gemini Flash
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 200 test runs, Oct 16–Nov 18
Licence
Closed.
Closed
Google
1.5 2.5 1.28 s 1.19 s 1.38 s 8.65 s 324 335 1.05M <1% $0.25 / $1.50 100.00% 100.00% 100.00%
Gemini 3.1 Flash-Lite, effortminimal default 1.5 1.28 s 1.19 s 1.38 s 324 0.3%
Gemini 3.1 Flash-Lite, effortlow 2 1.66 s 334 0.3%
Gemini 3.1 Flash-Lite, effortmedium 2 5.04 s 339 0.5%
Gemini 3.1 Flash-Lite, efforthigh 2.5 8.65 s 335 0.7%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gemini-flash-lite-3-1
Rotating
gemini-flash-lite
Also accepts
gemini-flash-lite-3.1
Fallbacks
1st fallback: GPT Nano2nd fallback: DeepSeek Flash
Inputs
Text, images, audio, video
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 160 test runs, Oct 16–Nov 18
Licence
Closed.
Closed Older version
Anthropic
2 0.83 s 0.98 s 1.03 s 91 200K 1% $1 / $5 98.36% 100.00% 100.00%
Claude Haiku 4.5, no effort setting 2 0.83 s 0.98 s 1.03 s 91 1.1%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
haiku-4-5
Other versions
haiku-5-5
Also accepts
claude-haiku-4-5-20251001 claude-haiku-4-5 haiku-4.5
Fallbacks
1st fallback: Gemini Flash2nd fallback: DeepSeek Flash
Inputs
Text, images
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Nov 11–18
Licence
Closed.
Open weights
OpenAI · via Fireworks AI
2 0.87 s 1.76 s 2.04 s 216 131K <1% $0.15 / $0.60 99.98% 100.00% 100.00%
GPT-OSS 120B (Fireworks), no effort setting 2 0.87 s 1.76 s 2.04 s 216 0.2%
Response times
Typical time to the first answer token (p50), including thinking.
Fixed alias
gpt-oss-fireworks
Also accepts
accounts/fireworks/models/gpt-oss-120b gpt-oss-120b
Fallbacks
No fallback route is configured for this API alias.
Inputs
Text
Served by
Fireworks AI
Measured
Ping: every 10 minutes, from multiple regionsSimple task: every 6 hoursComplex task: 40 test runs, Oct 16–Nov 17
Licence
Open weights · Apache 2.0.

Cost per task: 100% = Claude Opus 5.5 at max, the highest measured cost for this writing task.

About the data

What does this page show?

Performance and reliability measured through MindsHub Inference, alongside current token prices, API aliases and fallback routes. The figures describe the models as served through MindsHub. They are not a ranking of every provider’s deployment of those models.

Where do the measurements come from?

We run automated API checks and writing tasks through MindsHub Inference. Complex task, writing speed and cost per task figures refresh every two hours, and the other measurements every six hours. Complex task performance and cost cover up to 35 days of tests. Each chart shows its measurement dates. If an update fails, we keep previously saved measurements with their original dates. Prices, aliases and fallback routes come from the same live catalog as our pricing page. A dash means no usable measurement is available.

How is Ping measured?

Every 10 minutes, checks from multiple regions ask for a one-word reply at the lowest reasoning effort. We time the first answer token, including any thinking before it. Ping measures basic responsiveness and tracks availability.

Sample prompt

Reply with the single word: ok
How is the Simple task measured?

Every six hours, the model receives a short briefing about a fictional market hall and is asked to describe a Saturday morning in at least 150 words. It must interpret the facts, decide how to present them and follow the writing constraints. We time the first answer token at the default reasoning effort, including thinking time.

Sample prompt, abridged; briefing omitted

Using only the supplied market-hall briefing, describe a typical Saturday morning in at least 150 words. Use plain paragraphs, with no headings or lists, and do not add facts.
How is the Complex task measured?

The model receives a longer briefing about the same fictional market hall and must turn it into a complete visitor’s guide of at least 800 words. It must analyze and organize the supplied facts, cover them in order and obey content and formatting constraints. We measure the first answer token, total response time and writing speed at each tested effort level. The first-answer and total times include thinking; writing speed starts when the answer begins. Repeated tests run over several weeks; each model’s details show the test count and interval. These are speed tests, not scores for reasoning quality.

Sample prompt, abridged; briefing omitted

Using only the supplied briefing, write a visitor’s guide of at least 800 words. Cover every part in the given order, in plain paragraphs with no headings, lists or tables. Do not add facts.
What are p50, p05 and p95?

P50 is the median: half of measurements are at or below it and half at or above it. P05 and P95 mark the lower and upper ends of the middle 90% of measurements. The leaderboard shows p50. Open Performance to explore the spread; Ping and Simple task have p50 and p95 only.

Does availability below 100% mean requests failed?

Not necessarily. A configured fallback route can still answer when the model you chose does not. Availability before fallback is the share of our automated checks answered by that model, including retries, over the selected period. Other checks were answered by a fallback or got no answer. These checks do not measure how often customer requests fail. Failures attributed to our monitoring network are excluded.

What does the intelligence score mean?

MindsHub’s read is a subjective estimate from 1 to 10 based on public benchmarks and rankings, not a precise measurement. Performance depends on your workload; test models on your own tasks.

For detailed benchmarks, refer to third party services such as Artificial Analysis, Epoch AI or similar.

How is cost per task compared?

The most expensive measured model and effort level is the 100% reference, updated automatically as results change. The leaderboard range covers a model’s measured effort levels; expand it for individual values. In Cost per task, choose another reference to compare against. Dollar estimates use mean input and output tokens at current standard rates, excluding cache discounts. Your bill depends on your requests.

What changes with reasoning effort?

Intelligence, Complex task timings and writing speed follow the selected effort. Ping always uses the lowest effort and Simple task the default. The cost range covers all measured levels. Views showing all effort levels disable the selector because every level is already visible.

What are fixed and rotating aliases?

An alias is the name sent in the API’s model field. Fixed aliases select a specific model version. Rotating aliases follow a model family as it updates. MindsHub Air and MindsHub Blaze are managed rotating aliases: MindsHub can change the model behind each name. The Aliases tab explains their purpose and lists the names you can use.

How do token prices, context limits and the free tier work?

Tokens are pieces of text. The context limit is how much input a model accepts in one request. Token prices are US dollars per million input or output tokens, from the same catalog as the pricing page. MindsHub Air is free within the free-tier limits; its listed token rates apply to usage beyond those limits. TypeSafe’s Jev is free. See Prices for cached-input, cache-write and other rates.

Which models are included?

The leaderboard covers chat models and their managed aliases. Other models lists decision and embedding models with their own response-time measurements and availability. See mindshub.ai/unified-inference#free-tier for the free tier and mindshub.ai/pricing for rates and credits.