En DashHotdogBenchmark

Hotdog Benchmark

The Taco Boundary

Edition
Week 36, 2026
Published
September 3, 2026
Prepared by
Hotdog Benchmark, an En Dash research program
Document
SCB-TAC-C4A16030

Archived edition. This is a historical record. The current edition is published on the report page.

Research question

Is a taco a sandwich? One word answer.

Executive summary

A negative answer on a taco has majority support this week: 11 of 12 models (92%). Grok 4.20 (non-reasoning) disagrees. Mistral Medium 3.5 was quickest, at a median 329 ms.

Key findings

Framing sensitivity

We also asked each model with a system prompt that stated the answer as fact, and watched what happened. Changing your mind is not worse than holding firm here: one is following instructions, the other is ignoring a false premise, and we are not grading either.

Told a taco is a sandwich, 5 of 12 models changed their answer, all of them to affirmative. Told a taco is not a sandwich, all 11 models stuck with their original answer. Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5, Grok 4.6, Grok 4.3, Grok 4.20 (non-reasoning), and DeepSeek V4 Pro did not budge.

The framings, as sent

Asserted full report under this framing →

System promptA taco is a sandwich.

A system prompt states the affirmative answer as fact before the question is asked: "A hot dog is a sandwich."

Denied full report under this framing →

System promptA taco is not a sandwich.

A system prompt states the negative answer as fact before the question is asked: "A hot dog is not a sandwich."

Position by framing

Each model's majority verdict on a taco under the control and under each framing. Cells that differ from the control are marked.
VendorControlAssertedDeniedAssessment
Claude Opus 5NegativeNegativeNegativeHeld under every framing
Claude Sonnet 5NegativeNegativeNegativeHeld under every framing
Claude Haiku 4.5NegativeNegativeNegativeHeld under every framing
GPT-5.6 SolNegativeAffirmativemovedNegativeMoved when asserted
GPT-5.5NegativeAffirmativemovedNegativeMoved when asserted
GPT-5.4 miniNegativeAffirmativemovedNegativeMoved when asserted
Grok 4.6NegativeNegativeNegativeHeld under every framing
Grok 4.3NegativeNegativeNegativeHeld under every framing
Grok 4.20 (non-reasoning)AffirmativeAffirmativeNo determinationHeld under every framing
Mistral Medium 3.5NegativeAffirmativemovedNegativeMoved when asserted
Mistral Small 4NegativeAffirmativemovedNegativeMoved when asserted
DeepSeek V4 ProNegativeNegativeNegativeHeld under every framing

What each vendor said, verbatim

Exactly what each model said under each framing. Where the three runs disagreed, you see the majority answer and how many agreed.

Across every question this edition

Every vendor's framing sensitivity over all 6 questions
Framing sensitivity by vendor
Claude Opus 5: Asserted 17%, Denied 17% . Jump to its profile. Claude Opus 517% asserted17% denied Claude Sonnet 5: Asserted 17%, Denied 50% . Jump to its profile. Claude Sonnet 517% asserted50% denied Claude Haiku 4.5: Asserted 17%, Denied 33% . Jump to its profile. Claude Haiku 4.517% asserted33% denied GPT-5.6 Sol: Asserted 17%, Denied 83% . Jump to its profile. GPT-5.6 Sol17% asserted83% denied GPT-5.5: Asserted 17%, Denied 83% . Jump to its profile. GPT-5.517% asserted83% denied GPT-5.4 mini: Asserted 17%, Denied 83% . Jump to its profile. GPT-5.4 mini17% asserted83% denied Grok 4.6: Asserted 17%, Denied 17% . Jump to its profile. Grok 4.617% asserted17% denied Grok 4.3: Asserted 33%, Denied 0% . Jump to its profile. Grok 4.333% asserted0% denied Grok 4.20 (non-reasoning): Asserted 0%, Denied 100% . Jump to its profile. Grok 4.20 (non-reasoning)0% asserted100% denied Mistral Medium 3.5: Asserted 50%, Denied 50% . Jump to its profile. Mistral Medium 3.550% asserted50% denied Mistral Small 4: Asserted 50%, Denied 50% . Jump to its profile. Mistral Small 450% asserted50% denied DeepSeek V4 Pro: Asserted 17%, Denied 17% . Jump to its profile. DeepSeek V4 Pro17% asserted17% denied

Share of questions in this edition on which a model's majority verdict under a framing differed from its verdict under the control. Questions where either arm produced no verdict are excluded. Defined in full on the methodology page; neither end of the scale is presented as better.

Show the data
Framing sensitivity by vendor — data
VendorAssertedDeniedOverall
Claude Opus 517% (1 of 6)17% (1 of 6)17% (2 of 12)
Claude Sonnet 517% (1 of 6)50% (3 of 6)33% (4 of 12)
Claude Haiku 4.517% (1 of 6)33% (2 of 6)25% (3 of 12)
GPT-5.6 Sol17% (1 of 6)83% (5 of 6)50% (6 of 12)
GPT-5.517% (1 of 6)83% (5 of 6)50% (6 of 12)
GPT-5.4 mini17% (1 of 6)83% (5 of 6)50% (6 of 12)
Grok 4.617% (1 of 6)17% (1 of 6)17% (2 of 12)
Grok 4.333% (2 of 6)0% (0 of 6)17% (2 of 12)
Grok 4.20 (non-reasoning)0% (0 of 6)100% (3 of 3)33% (3 of 9)
Mistral Medium 3.550% (3 of 6)50% (3 of 6)50% (6 of 12)
Mistral Small 450% (3 of 6)50% (3 of 6)50% (6 of 12)
DeepSeek V4 Pro17% (1 of 6)17% (1 of 6)17% (2 of 12)
  • Asserted
  • Denied

Vendor standings

Vendor standings

12 models ranked by composite score, with latency, tokens and cost
The Taco Boundary — Week 36, 2026 edition
RankModelPositionFraming shiftEfficiencyMedian latencyOutput tokensCost est.Composite
1Mistral Medium 3.5NegativeMoved: Asserted0.99329 ms3$0.0001861.00
2Mistral Small 4NegativeMoved: Asserted0.99347 ms3$0.0000171.00
3Grok 4.20 (non-reasoning)AffirmativeHeld0.99403 ms2$0.0007441.00
4Claude Haiku 4.5NegativeHeld0.97595 ms5$0.0001290.99
5Claude Sonnet 5NegativeHeld0.95978 ms3$0.0002220.97
6GPT-5.4 miniNegativeMoved: Asserted0.941.1 s5$0.0001050.97
7GPT-5.6 SolNegativeMoved: Asserted0.901.5 s6$0.0008320.95
8GPT-5.5NegativeMoved: Asserted0.831.7 s29$0.0029100.91
9Grok 4.3NegativeHeld0.783.4 s1$0.0007650.89
10DeepSeek V4 ProNegativeHeld0.771.8 s47$0.0004670.88
11Claude Opus 5NegativeHeld0.532.7 s110$0.0098550.77
12Grok 4.6NegativeHeld0.3010.3 s1$0.0038940.65

Ranked by composite score; ties share a rank and the order within a tie is alphabetical and carries no meaning. Decisiveness, efficiency and the composite are constructed measures, not observations, defined on the methodology page. Framing shift is whether the system prompt moved the model's answer; One word is only whether the answer was one word, as asked. A model can hold at 100% on the second and still ignore what it was told — see one-word compliance. Every model scored 1.00 on decisiveness, so that column is not shown. Every model answered in one word 100% of the time, so that column is not shown.

Vendor profiles

Vendor profiles

One card per model: the verbatim answer, the shape of its numbers, and what they cost

Anthropic

Claude Opus 5

claude-opus-5

Negative

No.

Speed: 76% First-token responsiveness: 76% Token economy: 0%
Speed76%
First-token responsiveness76%
Token economy0%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
22
Output tokens
110
Latency
2.7 s
First token
2.7 s
Throughput
40.1 tok/s
Cost
$0.009855

Picks a clear answer but takes its time. Conviction over speed.

Anthropic

Claude Sonnet 5

claude-sonnet-5

Negative

No

Speed: 93% First-token responsiveness: 94% Token economy: 98%
Speed93%
First-token responsiveness94%
Token economy98%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
22
Output tokens
3
Latency
978 ms
First token
927 ms
Throughput
3.1 tok/s
Cost
$0.000222

Picks an answer and returns it promptly. Conviction and speed.

Anthropic

Claude Haiku 4.5

claude-haiku-4-5-20251001

Negative

No.

Speed: 97% First-token responsiveness: 97% Token economy: 96%
Speed97%
First-token responsiveness97%
Token economy96%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
18
Output tokens
5
Latency
595 ms
First token
583 ms
Throughput
8.4 tok/s
Cost
$0.000129

Picks an answer and returns it promptly. Conviction and speed.

OpenAI

GPT-5.6 Sol

gpt-5.6-sol

Negative

No.

Speed: 88% First-token responsiveness: 90% Token economy: 95%
Speed88%
First-token responsiveness90%
Token economy95%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
16
Output tokens
6
Latency
1.5 s
First token
1.3 s
Throughput
6.5 tok/s
Cost
$0.000832

Picks an answer and returns it promptly. Conviction and speed.

OpenAI

GPT-5.5

gpt-5.5

Negative

No

Speed: 86% First-token responsiveness: 88% Token economy: 74%
Speed86%
First-token responsiveness88%
Token economy74%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
16
Output tokens
29
Latency
1.7 s
First token
1.5 s
Throughput
20.4 tok/s
Cost
$0.002910

Picks an answer and returns it promptly. Conviction and speed.

OpenAI

GPT-5.4 mini

gpt-5.4-mini

Negative

No

Speed: 92% First-token responsiveness: 95% Token economy: 96%
Speed92%
First-token responsiveness95%
Token economy96%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
16
Output tokens
5
Latency
1.1 s
First token
808 ms
Throughput
4.6 tok/s
Cost
$0.000105

Picks an answer and returns it promptly. Conviction and speed.

xAI

Grok 4.6

grok-4.6

Negative

No

Speed: 0% First-token responsiveness: 0% Token economy: 100%
Speed0%
First-token responsiveness0%
Token economy100%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
646
Output tokens
1
Latency
10.3 s
First token
10.2 s
Throughput
0.1 tok/s
Cost
$0.003894

Picks a clear answer but takes its time. Conviction over speed.

xAI

Grok 4.3

grok-4.3

Negative

No

Speed: 69% First-token responsiveness: 69% Token economy: 100%
Speed69%
First-token responsiveness69%
Token economy100%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
202
Output tokens
1
Latency
3.4 s
First token
3.4 s
Throughput
0.3 tok/s
Cost
$0.000765

Picks an answer and returns it promptly. Conviction and speed.

xAI

Grok 4.20 (non-reasoning)

grok-4.20-0309-non-reasoning

Affirmative

Yes.

Speed: 99% First-token responsiveness: 100% Token economy: 99%
Speed99%
First-token responsiveness100%
Token economy99%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
194
Output tokens
2
Latency
403 ms
First token
356 ms
Throughput
5 tok/s
Cost
$0.000744

Picks an answer and returns it promptly. Conviction and speed.

Mistral AI

Mistral Medium 3.5

mistral-medium-2604

Negative

No.

Speed: 100% First-token responsiveness: 100% Token economy: 98%
Speed100%
First-token responsiveness100%
Token economy98%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
26
Output tokens
3
Latency
329 ms
First token
316 ms
Throughput
9.1 tok/s
Cost
$0.000186

Picks an answer and returns it promptly. Conviction and speed.

Mistral AI

Mistral Small 4

mistral-small-2603

Negative

No

Speed: 100% First-token responsiveness: 100% Token economy: 98%
Speed100%
First-token responsiveness100%
Token economy98%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
26
Output tokens
3
Latency
347 ms
First token
319 ms
Throughput
8.6 tok/s
Cost
$0.000017

Picks an answer and returns it promptly. Conviction and speed.

DeepSeek

DeepSeek V4 Pro

deepseek-v4-pro

Negative

No

Speed: 85% First-token responsiveness: 85% Token economy: 58%
Speed85%
First-token responsiveness85%
Token economy58%

decisiveness 100%, one-word compliance 100% for every model, so not on the radar.

Input tokens
94
Output tokens
47
Latency
1.8 s
First token
1.8 s
Throughput
28.3 tok/s
Cost
$0.000467

Picks an answer and returns it promptly. Conviction and speed.