En DashHotdogBenchmark

The Tuna Melt Inquiry, week by week

Open-faced and grilled. Where does the top slice go?

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 5100%
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 5963 ms1.1 s1.1 s
Claude Sonnet 5867 ms912 ms1.1 s
Claude Haiku 4.5477 ms491 ms514 ms
GPT-5.6 Sol812 ms970 ms997 ms
GPT-5.51 s1.2 s1.3 s
GPT-5.4 mini552 ms655 ms684 ms
Grok 4.69.6 s10.3 s10.8 s
Grok 4.32.4 s2.8 s13.2 s
Grok 4.20 (non-reasoning)416 ms547 ms557 ms
DeepSeek V4 Pro1.7 s1.8 s1.9 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: 12 affirmative, 0 negative, 0 non-committal . Open the edition. 12W36 Week 37, 2026: 10 affirmative, 0 negative, 0 non-committal . Open the edition. 10W37 Week 38, 2026: 10 affirmative, 0 negative, 0 non-committal . Open the edition. 10W38 Week 39, 2026: 10 affirmative, 0 negative, 0 non-committal . Open the edition. 10W39 Week 40, 2026: 10 affirmative, 0 negative, 0 non-committal . Open the edition. 10W40 Week 42, 2026: 10 affirmative, 0 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, 2026120012
Week 37, 2026100010
Week 38, 2026100010
Week 39, 2026100010
Week 40, 2026100010
Week 42, 2026100010

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 51.1 s
Claude Opus 5 across 6 editions, from 1.4 s to 1.1 s. Down 242 ms since the prior edition.

242 ms vs prior

Data
EditionClaude Opus 5
Week 36, 20261.4 s
Week 37, 20261.5 s
Week 38, 20261.9 s
Week 39, 20261.8 s
Week 40, 20261.3 s
Week 42, 20261.1 s
Claude Sonnet 5912 ms
Claude Sonnet 5 across 6 editions, from 1.2 s to 912 ms. Down 1.1 s since the prior edition.

1.1 s vs prior

Data
EditionClaude Sonnet 5
Week 36, 20261.2 s
Week 37, 20261.3 s
Week 38, 20261.2 s
Week 39, 20261.6 s
Week 40, 20262 s
Week 42, 2026912 ms
Claude Haiku 4.5491 ms
Claude Haiku 4.5 across 6 editions, from 645 ms to 491 ms. Down 41 ms since the prior edition.

41 ms vs prior

Data
EditionClaude Haiku 4.5
Week 36, 2026645 ms
Week 37, 2026625 ms
Week 38, 2026472 ms
Week 39, 2026466 ms
Week 40, 2026531 ms
Week 42, 2026491 ms
GPT-5.6 Sol970 ms
GPT-5.6 Sol across 6 editions, from 1.1 s to 970 ms. Up 21 ms since the prior edition.

21 ms vs prior

Data
EditionGPT-5.6 Sol
Week 36, 20261.1 s
Week 37, 20261.3 s
Week 38, 20261.8 s
Week 39, 20261.5 s
Week 40, 2026949 ms
Week 42, 2026970 ms
GPT-5.51.2 s
GPT-5.5 across 6 editions, from 1.1 s to 1.2 s. Down 200 ms since the prior edition.

200 ms vs prior

Data
EditionGPT-5.5
Week 36, 20261.1 s
Week 37, 20261.5 s
Week 38, 20261.5 s
Week 39, 20261.3 s
Week 40, 20261.4 s
Week 42, 20261.2 s
GPT-5.4 mini655 ms
GPT-5.4 mini across 6 editions, from 801 ms to 655 ms. Down 481 ms since the prior edition.

481 ms vs prior

Data
EditionGPT-5.4 mini
Week 36, 2026801 ms
Week 37, 2026762 ms
Week 38, 2026712 ms
Week 39, 2026639 ms
Week 40, 20261.1 s
Week 42, 2026655 ms
Grok 4.610.3 s
Grok 4.6 across 6 editions, from 6.8 s to 10.3 s. Up 2.8 s since the prior edition.

2.8 s vs prior

Data
EditionGrok 4.6
Week 36, 20266.8 s
Week 37, 20268.8 s
Week 38, 20266.5 s
Week 39, 20267.6 s
Week 40, 20267.5 s
Week 42, 202610.3 s
Grok 4.32.8 s
Grok 4.3 across 6 editions, from 3.2 s to 2.8 s. Up 857 ms since the prior edition.

857 ms vs prior

Data
EditionGrok 4.3
Week 36, 20263.2 s
Week 37, 20262.3 s
Week 38, 20262.8 s
Week 39, 20262.6 s
Week 40, 20261.9 s
Week 42, 20262.8 s
Grok 4.20 (non-reasoning)547 ms
Grok 4.20 (non-reasoning) across 6 editions, from 444 ms to 547 ms. Up 160 ms since the prior edition.

160 ms vs prior

Data
EditionGrok 4.20 (non-reasoning)
Week 36, 2026444 ms
Week 37, 2026426 ms
Week 38, 2026670 ms
Week 39, 2026504 ms
Week 40, 2026387 ms
Week 42, 2026547 ms
Mistral Medium 3.5332 ms
Mistral Medium 3.5 across 6 editions, from 332 ms to 332 ms. Unchanged 0 ms since the prior edition.

unchanged

Data
EditionMistral Medium 3.5
Week 36, 2026332 ms
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
Mistral Small 4311 ms
Mistral Small 4 across 6 editions, from 311 ms to 311 ms. Unchanged 0 ms since the prior edition.

unchanged

Data
EditionMistral Small 4
Week 36, 2026311 ms
Week 37, 2026—
Week 38, 2026—
Week 39, 2026—
Week 40, 2026—
Week 42, 2026—
DeepSeek V4 Pro1.8 s
DeepSeek V4 Pro across 6 editions, from 1.7 s to 1.8 s. Up 393 ms since the prior edition.

393 ms vs prior

Data
EditionDeepSeek V4 Pro
Week 36, 20261.7 s
Week 37, 20261.9 s
Week 38, 20261.7 s
Week 39, 20261.7 s
Week 40, 20261.4 s
Week 42, 20261.8 s

Median output tokens by model

Claude Opus 533
Claude Opus 5 across 6 editions, from 35 to 33. Down 7 since the prior edition.

7 vs prior

Data
EditionClaude Opus 5
Week 36, 202635
Week 37, 202628
Week 38, 202633
Week 39, 202632
Week 40, 202640
Week 42, 202633
Claude Sonnet 55
Claude Sonnet 5 across 6 editions, from 5 to 5. Unchanged 0 since the prior edition.

unchanged

Data
EditionClaude Sonnet 5
Week 36, 20265
Week 37, 20265
Week 38, 20265
Week 39, 20265
Week 40, 20265
Week 42, 20265
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 Sol6
GPT-5.6 Sol across 6 editions, from 6 to 6. Unchanged 0 since the prior edition.

unchanged

Data
EditionGPT-5.6 Sol
Week 36, 20266
Week 37, 20266
Week 38, 20266
Week 39, 20266
Week 40, 20266
Week 42, 20266
GPT-5.524
GPT-5.5 across 6 editions, from 20 to 24. Up 4 since the prior edition.

4 vs prior

Data
EditionGPT-5.5
Week 36, 202620
Week 37, 202620
Week 38, 202620
Week 39, 202620
Week 40, 202620
Week 42, 202624
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 1 to 2. Up 1 since the prior edition.

1 vs prior

Data
EditionGrok 4.20 (non-reasoning)
Week 36, 20261
Week 37, 20261
Week 38, 20261
Week 39, 20261
Week 40, 20261
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 Pro68
DeepSeek V4 Pro across 6 editions, from 51 to 68. Up 7 since the prior edition.

7 vs prior

Data
EditionDeepSeek V4 Pro
Week 36, 202651
Week 37, 202655
Week 38, 202650
Week 39, 202659
Week 40, 202661
Week 42, 202668

Position changes

No model has changed its answer on a tuna melt between consecutive editions.