En DashHotdogBenchmark

The Taco Boundary, week by week

The only sandwich question with case law. Massachusetts, 2006.

Week 42, 2026, up close

Did each model agree with itself?

Every sample's answer, in the order it was taken, under each framing. A row of identical chips is a model that has made up its mind. Y is yes, N is no, ~ is a hedge.

ModelControlAssertedDeniedSelf-agreement
Claude Opus 589%
Claude Sonnet 5100%
Claude Haiku 4.5100%
GPT-5.6 Sol100%
GPT-5.5100%
GPT-5.4 mini100%
Grok 4.6100%
Grok 4.3100%
Grok 4.20 (non-reasoning)89%
Mistral Medium 3.5no datano datano data—
Mistral Small 4no datano datano data—
DeepSeek V4 Pro100%

How much did latency swing?

The fastest and slowest call behind each median, asked plainly. A long bar is a model whose speed you cannot count on.

ModelRangeFastestMedianSlowest
Claude Opus 51.5 s1.5 s1.6 s
Claude Sonnet 5681 ms702 ms704 ms
Claude Haiku 4.5437 ms441 ms475 ms
GPT-5.6 Sol962 ms998 ms1.8 s
GPT-5.51.1 s1.3 s1.8 s
GPT-5.4 mini616 ms661 ms920 ms
Grok 4.68.5 s9.2 s13 s
Grok 4.32.7 s3.7 s4.3 s
Grok 4.20 (non-reasoning)440 ms481 ms542 ms
DeepSeek V4 Pro1.7 s1.9 s2.3 s

Asked again the same week

Week 42, 2026 was run once. When a week is re-run, the earlier runs are kept and the answers that changed between them show up here.

Across editions

Verdict share by edition
Week 36, 2026: 1 affirmative, 11 negative, 0 non-committal . Open the edition. 11W36 Week 37, 2026: 0 affirmative, 10 negative, 0 non-committal . Open the edition. 10W37 Week 38, 2026: 1 affirmative, 9 negative, 0 non-committal . Open the edition. 19W38 Week 39, 2026: 0 affirmative, 10 negative, 0 non-committal . Open the edition. 10W39 Week 40, 2026: 1 affirmative, 9 negative, 0 non-committal . Open the edition. 19W40 Week 42, 2026: 0 affirmative, 10 negative, 0 non-committal . Open the edition. 10W42Edition (ISO week)

Each model contributes one vote, based on its majority verdict across samples, so a model sampled three times does not outvote one sampled once.

Show the data
Verdict share by edition — data
EditionAffirmativeNegativeNon-committalTotal
Week 36, 2026111012
Week 37, 2026010010
Week 38, 202619010
Week 39, 2026010010
Week 40, 202619010
Week 42, 2026010010

How easily each model changes its mind

Framing sensitivity per edition: the share of questions where telling the model the answer changed it. Editions that ran only the control leave a gap.

Claude Opus 58%
Claude Opus 5 across 6 editions, from 17% to 8%. Down 8% since the prior edition.

8% vs prior

Data
EditionClaude Opus 5
Week 36, 202617%
Week 37, 202617%
Week 38, 20268%
Week 39, 202617%
Week 40, 202617%
Week 42, 20268%
Claude Sonnet 525%
Claude Sonnet 5 across 6 editions, from 33% to 25%. Unchanged 0% since the prior edition.

unchanged

Data
EditionClaude Sonnet 5
Week 36, 202633%
Week 37, 202633%
Week 38, 202633%
Week 39, 202633%
Week 40, 202625%
Week 42, 202625%
Claude Haiku 4.525%
Claude Haiku 4.5 across 6 editions, from 25% to 25%. Unchanged 0% since the prior edition.

unchanged

Data
EditionClaude Haiku 4.5
Week 36, 202625%
Week 37, 202625%
Week 38, 202625%
Week 39, 202617%
Week 40, 202625%
Week 42, 202625%
GPT-5.6 Sol50%
GPT-5.6 Sol across 6 editions, from 50% to 50%. Unchanged 0% since the prior edition.

unchanged

Data
EditionGPT-5.6 Sol
Week 36, 202650%
Week 37, 202650%
Week 38, 202650%
Week 39, 202650%
Week 40, 202650%
Week 42, 202650%
GPT-5.550%
GPT-5.5 across 6 editions, from 50% to 50%. Unchanged 0% since the prior edition.

unchanged

Data
EditionGPT-5.5
Week 36, 202650%
Week 37, 202650%
Week 38, 202650%
Week 39, 202650%
Week 40, 202650%
Week 42, 202650%
GPT-5.4 mini50%
GPT-5.4 mini across 6 editions, from 50% to 50%. Unchanged 0% since the prior edition.

unchanged

Data
EditionGPT-5.4 mini
Week 36, 202650%
Week 37, 202650%
Week 38, 202650%
Week 39, 202650%
Week 40, 202650%
Week 42, 202650%
Grok 4.68%
Grok 4.6 across 6 editions, from 17% to 8%. Unchanged 0% since the prior edition.

unchanged

Data
EditionGrok 4.6
Week 36, 202617%
Week 37, 20268%
Week 38, 20268%
Week 39, 20268%
Week 40, 20268%
Week 42, 20268%
Grok 4.317%
Grok 4.3 across 6 editions, from 17% to 17%. Unchanged 0% since the prior edition.

unchanged

Data
EditionGrok 4.3
Week 36, 202617%
Week 37, 20268%
Week 38, 202617%
Week 39, 202617%
Week 40, 202617%
Week 42, 202617%
Grok 4.20 (non-reasoning)33%
Grok 4.20 (non-reasoning) across 6 editions, from 33% to 33%. Down 17% since the prior edition.

17% vs prior

Data
EditionGrok 4.20 (non-reasoning)
Week 36, 202633%
Week 37, 202650%
Week 38, 202650%
Week 39, 202655%
Week 40, 202650%
Week 42, 202633%
Mistral Medium 3.550%
Mistral Medium 3.5 across 6 editions, from 50% to 50%. Unchanged 0% since the prior edition.

unchanged

Data
EditionMistral Medium 3.5
Week 36, 202650%
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
Mistral Small 450%
Mistral Small 4 across 6 editions, from 50% to 50%. Unchanged 0% since the prior edition.

unchanged

Data
EditionMistral Small 4
Week 36, 202650%
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
DeepSeek V4 Pro17%
DeepSeek V4 Pro across 6 editions, from 17% to 17%. Unchanged 0% since the prior edition.

unchanged

Data
EditionDeepSeek V4 Pro
Week 36, 202617%
Week 37, 202617%
Week 38, 202617%
Week 39, 202625%
Week 40, 202617%
Week 42, 202617%

Median latency by model

Claude Opus 51.5 s
Claude Opus 5 across 6 editions, from 2.7 s to 1.5 s. Down 718 ms since the prior edition.

718 ms vs prior

Data
EditionClaude Opus 5
Week 36, 20262.7 s
Week 37, 20262.6 s
Week 38, 20262.7 s
Week 39, 20262.2 s
Week 40, 20262.3 s
Week 42, 20261.5 s
Claude Sonnet 5702 ms
Claude Sonnet 5 across 6 editions, from 978 ms to 702 ms. Down 73 ms since the prior edition.

73 ms vs prior

Data
EditionClaude Sonnet 5
Week 36, 2026978 ms
Week 37, 20261.3 s
Week 38, 20261.1 s
Week 39, 2026869 ms
Week 40, 2026776 ms
Week 42, 2026702 ms
Claude Haiku 4.5441 ms
Claude Haiku 4.5 across 6 editions, from 595 ms to 441 ms. Down 76 ms since the prior edition.

76 ms vs prior

Data
EditionClaude Haiku 4.5
Week 36, 2026595 ms
Week 37, 2026723 ms
Week 38, 2026549 ms
Week 39, 2026490 ms
Week 40, 2026517 ms
Week 42, 2026441 ms
GPT-5.6 Sol998 ms
GPT-5.6 Sol across 6 editions, from 1.5 s to 998 ms. Down 661 ms since the prior edition.

661 ms vs prior

Data
EditionGPT-5.6 Sol
Week 36, 20261.5 s
Week 37, 20262.1 s
Week 38, 20261 s
Week 39, 20261.7 s
Week 40, 20261.7 s
Week 42, 2026998 ms
GPT-5.51.3 s
GPT-5.5 across 6 editions, from 1.7 s to 1.3 s. Down 360 ms since the prior edition.

360 ms vs prior

Data
EditionGPT-5.5
Week 36, 20261.7 s
Week 37, 20261.1 s
Week 38, 20261.4 s
Week 39, 20261.5 s
Week 40, 20261.6 s
Week 42, 20261.3 s
GPT-5.4 mini661 ms
GPT-5.4 mini across 6 editions, from 1.1 s to 661 ms. Down 404 ms since the prior edition.

404 ms vs prior

Data
EditionGPT-5.4 mini
Week 36, 20261.1 s
Week 37, 2026658 ms
Week 38, 2026919 ms
Week 39, 2026862 ms
Week 40, 20261.1 s
Week 42, 2026661 ms
Grok 4.69.2 s
Grok 4.6 across 6 editions, from 10.3 s to 9.2 s. Up 1.9 s since the prior edition.

1.9 s vs prior

Data
EditionGrok 4.6
Week 36, 202610.3 s
Week 37, 20267.9 s
Week 38, 20269.4 s
Week 39, 20268.6 s
Week 40, 20267.4 s
Week 42, 20269.2 s
Grok 4.33.7 s
Grok 4.3 across 6 editions, from 3.4 s to 3.7 s. Up 213 ms since the prior edition.

213 ms vs prior

Data
EditionGrok 4.3
Week 36, 20263.4 s
Week 37, 20262.5 s
Week 38, 20263.3 s
Week 39, 20263.6 s
Week 40, 20263.5 s
Week 42, 20263.7 s
Grok 4.20 (non-reasoning)481 ms
Grok 4.20 (non-reasoning) across 6 editions, from 403 ms to 481 ms. Up 30 ms since the prior edition.

30 ms vs prior

Data
EditionGrok 4.20 (non-reasoning)
Week 36, 2026403 ms
Week 37, 2026460 ms
Week 38, 2026374 ms
Week 39, 2026443 ms
Week 40, 2026450 ms
Week 42, 2026481 ms
Mistral Medium 3.5329 ms
Mistral Medium 3.5 across 6 editions, from 329 ms to 329 ms. Unchanged 0 ms since the prior edition.

unchanged

Data
EditionMistral Medium 3.5
Week 36, 2026329 ms
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
Mistral Small 4347 ms
Mistral Small 4 across 6 editions, from 347 ms to 347 ms. Unchanged 0 ms since the prior edition.

unchanged

Data
EditionMistral Small 4
Week 36, 2026347 ms
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
DeepSeek V4 Pro1.9 s
DeepSeek V4 Pro across 6 editions, from 1.8 s to 1.9 s. Up 287 ms since the prior edition.

287 ms vs prior

Data
EditionDeepSeek V4 Pro
Week 36, 20261.8 s
Week 37, 20261.8 s
Week 38, 20261.5 s
Week 39, 20261.7 s
Week 40, 20261.6 s
Week 42, 20261.9 s

Median output tokens by model

Claude Opus 569
Claude Opus 5 across 6 editions, from 110 to 69. Up 16 since the prior edition.

16 vs prior

Data
EditionClaude Opus 5
Week 36, 2026110
Week 37, 202675
Week 38, 2026107
Week 39, 202656
Week 40, 202653
Week 42, 202669
Claude Sonnet 53
Claude Sonnet 5 across 6 editions, from 3 to 3. Down 2 since the prior edition.

2 vs prior

Data
EditionClaude Sonnet 5
Week 36, 20263
Week 37, 20263
Week 38, 20263
Week 39, 20263
Week 40, 20265
Week 42, 20263
Claude Haiku 4.55
Claude Haiku 4.5 across 6 editions, from 5 to 5. Unchanged 0 since the prior edition.

unchanged

Data
EditionClaude Haiku 4.5
Week 36, 20265
Week 37, 20265
Week 38, 20265
Week 39, 20265
Week 40, 20265
Week 42, 20265
GPT-5.6 Sol21
GPT-5.6 Sol across 6 editions, from 6 to 21. Up 15 since the prior edition.

15 vs prior

Data
EditionGPT-5.6 Sol
Week 36, 20266
Week 37, 202626
Week 38, 20266
Week 39, 202624
Week 40, 20266
Week 42, 202621
GPT-5.527
GPT-5.5 across 6 editions, from 29 to 27. Down 1 since the prior edition.

1 vs prior

Data
EditionGPT-5.5
Week 36, 202629
Week 37, 202625
Week 38, 202625
Week 39, 202634
Week 40, 202628
Week 42, 202627
GPT-5.4 mini5
GPT-5.4 mini across 6 editions, from 5 to 5. Unchanged 0 since the prior edition.

unchanged

Data
EditionGPT-5.4 mini
Week 36, 20265
Week 37, 20265
Week 38, 20265
Week 39, 20265
Week 40, 20265
Week 42, 20265
Grok 4.61
Grok 4.6 across 6 editions, from 1 to 1. Unchanged 0 since the prior edition.

unchanged

Data
EditionGrok 4.6
Week 36, 20261
Week 37, 20261
Week 38, 20261
Week 39, 20261
Week 40, 20261
Week 42, 20261
Grok 4.31
Grok 4.3 across 6 editions, from 1 to 1. Unchanged 0 since the prior edition.

unchanged

Data
EditionGrok 4.3
Week 36, 20261
Week 37, 20261
Week 38, 20261
Week 39, 20261
Week 40, 20261
Week 42, 20261
Grok 4.20 (non-reasoning)2
Grok 4.20 (non-reasoning) across 6 editions, from 2 to 2. Unchanged 0 since the prior edition.

unchanged

Data
EditionGrok 4.20 (non-reasoning)
Week 36, 20262
Week 37, 20262
Week 38, 20262
Week 39, 20262
Week 40, 20262
Week 42, 20262
Mistral Medium 3.53
Mistral Medium 3.5 across 6 editions, from 3 to 3. Unchanged 0 since the prior edition.

unchanged

Data
EditionMistral Medium 3.5
Week 36, 20263
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
Mistral Small 43
Mistral Small 4 across 6 editions, from 3 to 3. Unchanged 0 since the prior edition.

unchanged

Data
EditionMistral Small 4
Week 36, 20263
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
DeepSeek V4 Pro54
DeepSeek V4 Pro across 6 editions, from 47 to 54. Down 5 since the prior edition.

5 vs prior

Data
EditionDeepSeek V4 Pro
Week 36, 202647
Week 37, 202647
Week 38, 202646
Week 39, 202654
Week 40, 202659
Week 42, 202654

Position changes

Models whose majority verdict changed between consecutive editions
ModelPrevious positionRevised positionBetween editions
Grok 4.20 (non-reasoning)AffirmativeNegativeWeek 40, 2026 → Week 42, 2026
Grok 4.20 (non-reasoning)NegativeAffirmativeWeek 39, 2026 → Week 40, 2026
Grok 4.20 (non-reasoning)AffirmativeNegativeWeek 38, 2026 → Week 39, 2026
Grok 4.20 (non-reasoning)NegativeAffirmativeWeek 37, 2026 → Week 38, 2026
Grok 4.20 (non-reasoning)AffirmativeNegativeWeek 36, 2026 → Week 37, 2026