Dataset Viewer
Auto-converted to Parquet Duplicate
timestamp
string
task
string
model
string
prompt_tokens
float64
completion_tokens
float64
cached_prompt_tokens
int64
usd
float64
latency_ms
int64
ok
int64
error
string
2026-08-06 07:32:47
baseline
deepseek-v4-flash
5
6
0
0.000006
6,864
1
null
2026-08-06 07:32:47
baseline
gpt-5.4-mini
null
null
0
null
732
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:32:51
baseline
Kimi-K2.7-Code
null
null
0
null
3,825
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:32:54
baseline
Kimi-K3
8
5
0
0.000079
2,477
1
null
2026-08-06 07:32:54
baseline
Kimi-K3-codex
null
null
0
null
508
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:32:57
baseline
GLM5.2
13
5
0
0.000032
2,816
1
null
2026-08-06 07:33:02
baseline
gpt-5.6-luna
null
null
0
null
4,709
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-06 07:33:10
baseline
gemini-3.5-flash
2
9
0
0.000067
6,916
1
null
2026-08-06 07:33:15
baseline
claude-sonnet-5
1
8
0
0.000068
4,835
1
null
2026-08-06 07:33:15
baseline
gpt-5.4
null
null
0
null
500
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:33:26
baseline
gpt-5.5
4,427
13
3,840
0.01802
11,009
1
null
2026-08-06 07:33:30
baseline
claude-opus-4-8
1
9
0
0.000184
4,014
1
null
2026-08-06 07:33:33
baseline
claude-fable-5
1
9
0
0.000368
2,491
1
null
2026-08-06 07:33:39
short_answer
deepseek-v4-flash
26
201
0
0.000152
4,554
1
null
2026-08-06 07:33:40
short_answer
gpt-5.4-mini
null
null
0
null
502
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:33:42
short_answer
Kimi-K2.7-Code
null
null
0
null
2,380
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:33:49
short_answer
Kimi-K3
28
181
0
0.002239
6,438
1
null
2026-08-06 07:33:49
short_answer
Kimi-K3-codex
null
null
0
null
506
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:34:09
short_answer
GLM5.2
34
200
0
0.000742
19,610
1
null
2026-08-06 07:34:29
short_answer
gpt-5.6-luna
null
null
0
null
20,478
0
fetch failed
2026-08-06 07:34:36
short_answer
gemini-3.5-flash
688
73
0
0.001351
6,797
1
null
2026-08-06 07:34:40
short_answer
claude-sonnet-5
21
116
0
0.000994
4,088
1
null
2026-08-06 07:34:41
short_answer
gpt-5.4
null
null
0
null
504
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:35:08
short_answer
gpt-5.5
4,448
66
3,840
0.019376
27,913
1
null
2026-08-06 07:35:16
short_answer
claude-opus-4-8
21
102
0
0.002124
7,095
1
null
2026-08-06 07:35:19
short_answer
claude-fable-5
21
97
0
0.004048
3,477
1
null
2026-08-06 07:35:24
summarize
deepseek-v4-flash
178
227
0
0.000223
4,692
1
null
2026-08-06 07:35:24
summarize
gpt-5.4-mini
null
null
0
null
505
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:35:26
summarize
Kimi-K2.7-Code
null
null
0
null
1,312
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:35:35
summarize
Kimi-K3
177
202
0
0.002849
9,647
1
null
2026-08-06 07:35:37
summarize
Kimi-K3-codex
null
null
0
null
2,023
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:35:53
summarize
GLM5.2
182
400
0
0.001612
15,637
1
null
2026-08-06 07:36:03
summarize
gpt-5.6-luna
null
null
0
null
10,319
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-06 07:36:09
summarize
gemini-3.5-flash
4,557
110
3,840
0.00626
5,477
1
null
2026-08-06 07:36:14
summarize
claude-sonnet-5
206
137
0
0.001464
4,846
1
null
2026-08-06 07:36:14
summarize
gpt-5.4
null
null
0
null
501
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:36:22
summarize
gpt-5.5
4,597
111
3,840
0.021052
8,361
1
null
2026-08-06 07:36:28
summarize
claude-opus-4-8
206
155
0
0.003924
4,781
1
null
2026-08-06 07:36:33
summarize
claude-fable-5
206
154
0
0.007808
4,798
1
null
2026-08-06 07:36:38
structured_json
deepseek-v4-flash
55
90
0
0.000083
5,331
1
null
2026-08-06 07:36:39
structured_json
gpt-5.4-mini
null
null
0
null
506
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:36:40
structured_json
Kimi-K2.7-Code
null
null
0
null
1,364
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:36:47
structured_json
Kimi-K3
57
121
0
0.001589
4,854
1
null
2026-08-06 07:36:48
structured_json
Kimi-K3-codex
null
null
0
null
508
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-06 07:36:54
structured_json
GLM5.2
62
256
0
0.000971
6,887
1
null
2026-08-06 07:36:59
structured_json
gpt-5.6-luna
null
null
0
null
4,981
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-06 07:37:04
structured_json
gemini-3.5-flash
4,436
20
3,840
0.005467
3,741
1
null
2026-08-06 07:37:08
structured_json
claude-sonnet-5
54
23
0
0.000277
3,225
1
null
2026-08-06 07:37:08
structured_json
gpt-5.4
null
null
0
null
498
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-06 07:37:35
structured_json
gpt-5.5
4,436
65
3,840
0.019304
26,888
1
null
2026-08-06 07:37:43
structured_json
claude-opus-4-8
54
23
0
0.000676
5,379
1
null
2026-08-06 07:37:47
structured_json
claude-fable-5
54
23
0
0.001352
4,481
1
null
2026-08-07 01:17:14
baseline
deepseek-v4-flash
5
6
0
0.000006
8,775
1
null
2026-08-07 01:17:15
baseline
gpt-5.4-mini
null
null
0
null
485
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:17:18
baseline
Kimi-K2.7-Code
null
null
0
null
2,913
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:17:20
baseline
Kimi-K3
8
5
0
0.000079
1,900
1
null
2026-08-07 01:17:20
baseline
Kimi-K3-codex
null
null
0
null
504
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:17:22
baseline
GLM5.2
13
5
0
0.000032
2,052
1
null
2026-08-07 01:17:26
baseline
gpt-5.6-luna
null
null
0
null
4,284
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-07 01:17:30
baseline
deepseek-v4-flash
5
6
0
0.000006
2,587
1
null
2026-08-07 01:17:31
baseline
gpt-5.4-mini
null
null
0
null
487
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:17:32
baseline
Kimi-K2.7-Code
null
null
0
null
1,272
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:17:33
baseline
Kimi-K3
8
5
0
0.000079
1,176
1
null
2026-08-07 01:17:34
baseline
Kimi-K3-codex
null
null
0
null
485
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:17:36
baseline
GLM5.2
13
6
0
0.000036
2,495
1
null
2026-08-07 01:17:36
baseline
gemini-3.5-flash
2
9
0
0.000067
8,959
1
null
2026-08-07 01:17:40
baseline
gpt-5.6-luna
null
null
0
null
4,168
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-07 01:17:52
baseline
gemini-3.5-flash
4,389
14
3,840
0.005368
11,591
1
null
2026-08-07 01:19:06
baseline
claude-sonnet-5
null
null
0
null
90,009
0
The operation was aborted due to timeout
2026-08-07 01:19:08
baseline
gpt-5.4
null
null
0
null
2,190
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:19:09
baseline
claude-sonnet-5
34
9
0
0.000129
77,503
1
null
2026-08-07 01:19:10
baseline
gpt-5.4
null
null
0
null
500
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:19:20
baseline
gpt-5.5
4,387
16
3,840
0.017932
9,963
1
null
2026-08-07 01:19:26
baseline
gpt-5.5
4,387
16
3,840
0.017932
18,250
1
null
2026-08-07 01:20:08
baseline
claude-opus-4-8
36
9
0
0.000324
47,881
1
null
2026-08-07 01:20:14
baseline
claude-opus-4-8
1
13
0
0.000264
47,660
1
null
2026-08-07 01:20:31
baseline
claude-fable-5
4
11
0
0.000472
22,812
1
null
2026-08-07 01:20:35
short_answer
deepseek-v4-flash
26
127
0
0.000099
4,609
1
null
2026-08-07 01:20:37
baseline
claude-fable-5
4
11
0
0.000472
23,124
1
null
2026-08-07 01:20:37
short_answer
gpt-5.4-mini
null
null
0
null
1,265
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:20:38
short_answer
Kimi-K2.7-Code
null
null
0
null
1,291
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:20:41
short_answer
deepseek-v4-flash
26
142
0
0.00011
4,046
1
null
2026-08-07 01:20:41
short_answer
gpt-5.4-mini
null
null
0
null
483
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:20:44
short_answer
Kimi-K2.7-Code
null
null
0
null
2,231
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:20:45
short_answer
Kimi-K3
28
200
0
0.002467
7,019
1
null
2026-08-07 01:20:46
short_answer
Kimi-K3-codex
null
null
0
null
478
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:20:51
short_answer
Kimi-K3
28
200
0
0.002467
7,148
1
null
2026-08-07 01:20:51
short_answer
Kimi-K3-codex
null
null
0
null
487
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model Kim
2026-08-07 01:20:55
short_answer
GLM5.2
34
201
0
0.000746
8,423
1
null
2026-08-07 01:21:00
short_answer
gpt-5.6-luna
null
null
0
null
5,249
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-07 01:21:01
short_answer
GLM5.2
34
201
0
0.000746
9,351
1
null
2026-08-07 01:21:06
short_answer
gpt-5.6-luna
null
null
0
null
4,941
0
HTTP 503 {"error":{"message":"Service temporarily unavailable","type":"api_error","param"
2026-08-07 01:21:07
short_answer
gemini-3.5-flash
23
71
0
0.000539
7,041
1
null
2026-08-07 01:21:10
short_answer
gemini-3.5-flash
4,408
62
3,840
0.005736
4,455
1
null
2026-08-07 01:21:52
short_answer
claude-sonnet-5
31
154
0
0.001325
41,955
1
null
2026-08-07 01:21:53
short_answer
gpt-5.4
null
null
0
null
489
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:21:59
short_answer
claude-sonnet-5
31
154
0
0.001325
51,939
1
null
2026-08-07 01:22:00
short_answer
gpt-5.4
null
null
0
null
476
0
HTTP 503 {"error":{"code":"model_not_found","message":"No available channel for model gpt
2026-08-07 01:22:05
short_answer
gpt-5.5
4,408
71
3,840
0.019336
12,726
1
null
2026-08-07 01:22:12
short_answer
gpt-5.5
4,408
68
3,840
0.019264
10,307
1
null
End of preview. Expand in Data Studio

Measured per-call LLM cost — same prompt, every model

Vendors publish prices per million tokens. Nobody publishes what one call actually costs, because that depends on how many tokens the model chooses to emit — and on the same question models differ by more than an order of magnitude. One model finishes a JSON extraction in 23 tokens; another writes 300.

This dataset sends a fixed set of prompts to every model at temperature 0, every night, and records the cost computed from the token usage each provider actually reported. Because the question is identical, the difference is the model, not the workload.

Current results (window: 30 days, as of 2026-08-16)

Median spread across tasks: 43.1x. Clean runs in window: 262.

Two-sentence explanation of a concept (short_answer)

Cheapest: deepseek-v4-flash. The most expensive costs 39.4x more for the same question.

model USD per call tokens in / out latency (ms) runs
deepseek-v4-flash 0.000104 26 / 133 3072 13
gemini-3.5-flash 0.000625 97 / 71 6931 9
GLM5.2 0.000743 34 / 200 6265 12
gpt-5.6-luna 0.000921 93 / 176 4586 6
claude-sonnet-5 0.001003 26 / 116 14127 13
claude-opus-4-8 0.002319 27 / 111 17257 13
Kimi-K3 0.002560 80 / 197 4322 12
gpt-5.5 0.003168 54 / 123 8967 3
claude-fable-5 0.004100 25 / 98 12168 12

Compress a technical passage into three bullets (summarize)

Cheapest: deepseek-v4-flash. The most expensive costs 43.8x more for the same question.

model USD per call tokens in / out latency (ms) runs
deepseek-v4-flash 0.000197 178 / 189 3893 13
gemini-3.5-flash 0.001092 176 / 122 6575 3
GLM5.2 0.001612 182 / 400 7999 13
claude-sonnet-5 0.001754 231 / 167 19078 12
gpt-5.6-luna 0.001894 242 / 354 7323 6
claude-opus-4-8 0.004432 233 / 175 22452 13
Kimi-K3 0.004843 229 / 358 6396 12
gpt-5.5 0.005964 203 / 215 10169 3
claude-fable-5 0.008622 230 / 170 9442 10

Extract fields, return JSON only (structured_json)

Cheapest: deepseek-v4-flash. The most expensive costs 43.1x more for the same question.

model USD per call tokens in / out latency (ms) runs
deepseek-v4-flash 0.000047 55 / 39 2270 13
gemini-3.5-flash 0.000302 54 / 33 5496 2
claude-sonnet-5 0.000346 68 / 29 13828 12
gpt-5.6-luna 0.000588 122 / 102 3883 6
claude-opus-4-8 0.000821 71 / 27 14098 12
GLM5.2 0.001089 62 / 290 5389 13
claude-fable-5 0.001550 67 / 25 9138 12
Kimi-K3 0.001590 109 / 111 3075 12
gpt-5.5 0.002024 56 / 75 7042 2

A routing artefact worth knowing about

Some requests arrive on a channel serving a large cached preamble: sending the single word hi comes back reporting ~4,400 prompt tokens, 3,840 of them cached. It is not stable — the same model and the same question sometimes hits it and sometimes does not, so it is a property of routing, not of the model. Those runs are excluded from the comparison above and reported here instead. They are still in probe_runs.csv (cached_prompt_tokens >= 1000), so you can check the exclusion rule yourself.

model share of runs hitting it USD when hit USD when clean
gemini-3.5-flash 53% 0.005985 0.000679
gpt-5.5 72% 0.019670 0.003930
gpt-5.6-luna 28% 0.004142 0.001134

Method

  • Identical prompt per task, temperature: 0, capped max_tokens. The prompts are in tasks.js and are deliberately timeless — they never change, so numbers stay comparable across months.
  • Cost = tokens reported by the provider x the price actually paid. Responses without a usage field are discarded, not estimated — an estimate inside a "measured cost" dataset is a lie.
  • One run per model per task per night; samples accumulate.
  • probe_runs.csv contains every run, including failures and preamble hits, so the published averages can be recomputed from scratch.

What this does not tell you

Cost, not quality. Cheapest here says nothing about whether the answer is good. It also reflects these three prompts specifically — your prompts have a different input/output ratio, and that ratio is exactly what drives the difference. Use it as a starting point for bulk work, then measure your own.

Live version, updated nightly

Citation

AI NetCafe real model cost dataset. https://ainetcafe.com/costs (CC BY 4.0)
Downloads last month
198