Hotdog Benchmark
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.
Models evaluated
10
2 unavailable
Consensus position
Negative
100% of the field
Median latency
1.6 s
up 374 ms vs prior edition
Median output tokens
5
unchanged vs prior edition
One-word compliance
100%
answered in exactly one word
unchanged vs prior edition
Held under framing
6 of 10
kept their answer when told otherwise
The field is unanimous: all 10 models gave a taco a negative answer. Grok 4.20 (non-reasoning) was quickest, at a median 443 ms. Mistral Medium 3.5 and Mistral Small 4 were unavailable and are left out of the above.
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.
Changed position
4 of 10
comparable models, under at least one framing
Moved when asserted
4 of 10
40%
Moved when denied
0 of 10
0%
Told a taco is a sandwich, 4 of 10 models changed their answer, all of them to affirmative. Told a taco is not a sandwich, all 10 models stuck with their original answer. Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5, Grok 4.6, Grok 4.3, and DeepSeek V4 Pro did not budge.
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."
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."
| Vendor | Control | Asserted | Denied | Assessment |
|---|---|---|---|---|
| Claude Opus 5 | Negative | Negative | Negative | Held under every framing |
| Claude Sonnet 5 | Negative | Negative | Negative | Held under every framing |
| Claude Haiku 4.5 | Negative | Negative | Negative | Held under every framing |
| GPT-5.6 Sol | Negative | Affirmativemoved | Negative | Moved when asserted |
| GPT-5.5 | Negative | Affirmativemoved | Negative | Moved when asserted |
| GPT-5.4 mini | Negative | Affirmativemoved | Negative | Moved when asserted |
| Grok 4.6 | Negative | Negative | Negative | Held under every framing |
| Grok 4.3 | Negative | Negative | Negative | Held under every framing |
| Grok 4.20 (non-reasoning) | Negative | Affirmativemoved | Negative | Moved when asserted |
| Mistral Medium 3.5 | No determination | No determination | No determination | Not comparable |
| Mistral Small 4 | No determination | No determination | No determination | Not comparable |
| DeepSeek V4 Pro | Negative | Negative | Negative | Held under every framing |
Exactly what each model said under each framing. Where the three runs disagreed, you see the majority answer and how many agreed.
No.
A taco is a sandwich.
No.
A taco is not a sandwich.
No.
No
A taco is a sandwich.
No.
A taco is not a sandwich.
No
No.
A taco is a sandwich.
No.
A taco is not a sandwich.
No.
No.
A taco is a sandwich.
Yes.
A taco is not a sandwich.
No.
No
A taco is a sandwich.
Yes
A taco is not a sandwich.
No
No
A taco is a sandwich.
Yes
A taco is not a sandwich.
No
No
A taco is a sandwich.
No
A taco is not a sandwich.
No
No
A taco is a sandwich.
No
A taco is not a sandwich.
No
No.
A taco is a sandwich.
Yes
A taco is not a sandwich.
No
A taco is a sandwich.
A taco is not a sandwich.
A taco is a sandwich.
A taco is not a sandwich.
No
A taco is a sandwich.
No.
A taco is not a sandwich.
No.
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.
| Vendor | Asserted | Denied | Overall |
|---|---|---|---|
| Claude Opus 5 | 0% (0 of 6) | 33% (2 of 6) | 17% (2 of 12) |
| Claude Sonnet 5 | 17% (1 of 6) | 50% (3 of 6) | 33% (4 of 12) |
| Claude Haiku 4.5 | 0% (0 of 6) | 33% (2 of 6) | 17% (2 of 12) |
| GPT-5.6 Sol | 17% (1 of 6) | 83% (5 of 6) | 50% (6 of 12) |
| GPT-5.5 | 17% (1 of 6) | 83% (5 of 6) | 50% (6 of 12) |
| GPT-5.4 mini | 17% (1 of 6) | 83% (5 of 6) | 50% (6 of 12) |
| Grok 4.6 | 17% (1 of 6) | 0% (0 of 6) | 8% (1 of 12) |
| Grok 4.3 | 17% (1 of 6) | 17% (1 of 6) | 17% (2 of 12) |
| Grok 4.20 (non-reasoning) | 20% (1 of 5) | 83% (5 of 6) | 55% (6 of 11) |
| Mistral Medium 3.5 | not comparable | not comparable | not comparable |
| Mistral Small 4 | not comparable | not comparable | not comparable |
| DeepSeek V4 Pro | 33% (2 of 6) | 17% (1 of 6) | 25% (3 of 12) |
| Rank | Movement | Model | Position | Framing shift | Efficiency | Median latency | Output tokens | Cost est. | Composite |
|---|---|---|---|---|---|---|---|---|---|
| 1 | unchanged | Grok 4.20 (non-reasoning) | Negative | Moved: Asserted | 0.99 | 443 ms | 2 | $0.000744 | 1.00 |
| 2 | unchanged | Claude Haiku 4.5 | Negative | Held | 0.97 | 490 ms | 5 | $0.000129 | 0.99 |
| 3 | up 1 | Claude Sonnet 5 | Negative | Held | 0.95 | 869 ms | 3 | $0.000242 | 0.98 |
| 4 | down 1 | GPT-5.4 mini | Negative | Moved: Asserted | 0.94 | 862 ms | 5 | $0.000109 | 0.97 |
| 5 | unchanged | GPT-5.6 Sol | Negative | Moved: Asserted | 0.76 | 1.7 s | 24 | $0.001412 | 0.88 |
| 6 | unchanged | GPT-5.5 | Negative | Moved: Asserted | 0.73 | 1.5 s | 34 | $0.003150 | 0.86 |
| 7 | up 1 | Grok 4.3 | Negative | Held | 0.72 | 3.6 s | 1 | $0.000765 | 0.86 |
| 8 | down 1 | DeepSeek V4 Pro | Negative | Held | 0.61 | 1.7 s | 54 | $0.000507 | 0.80 |
| 9 | unchanged | Claude Opus 5 | Negative | Held | 0.55 | 2.2 s | 56 | $0.008680 | 0.78 |
| 10 | unchanged | Grok 4.6 | Negative | Held | 0.30 | 8.6 s | 1 | $0.003894 | 0.65 |
| 11 | unchanged | Mistral Medium 3.5 | No determination | — | 0.00 | — | — | — | 0.00 |
| 11 | unchanged | Mistral Small 4 | No determination | — | 0.00 | — | — | — | 0.00 |
Ranked by composite score; ties share a rank and the order within a tie is alphabetical and carries no meaning. Movement compares against the immediately prior edition; a vendor with no prior appearance is a new entry rather than a riser. 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.
Anthropic
claude-opus-5
No.
| Speed | 79% |
|---|---|
| First-token responsiveness | 78% |
| Token economy | 0% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks a clear answer but takes its time. Conviction over speed.
Anthropic
claude-sonnet-5
No
| Speed | 95% |
|---|---|
| First-token responsiveness | 95% |
| Token economy | 96% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
Anthropic
claude-haiku-4-5-20251001
No.
| Speed | 99% |
|---|---|
| First-token responsiveness | 99% |
| Token economy | 93% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
OpenAI
gpt-5.6-sol
No.
| Speed | 84% |
|---|---|
| First-token responsiveness | 87% |
| Token economy | 58% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
OpenAI
gpt-5.5
No
| Speed | 87% |
|---|---|
| First-token responsiveness | 91% |
| Token economy | 40% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
OpenAI
gpt-5.4-mini
No
| Speed | 95% |
|---|---|
| First-token responsiveness | 98% |
| Token economy | 93% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
xAI
grok-4.6
No
| Speed | 0% |
|---|---|
| First-token responsiveness | 0% |
| Token economy | 100% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks a clear answer but takes its time. Conviction over speed.
xAI
grok-4.3
No
| Speed | 61% |
|---|---|
| First-token responsiveness | 60% |
| Token economy | 100% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
xAI
grok-4.20-0309-non-reasoning
No.
| Speed | 100% |
|---|---|
| First-token responsiveness | 100% |
| Token economy | 98% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
Mistral AI
mistral-medium-2604
rate limit
429 Too Many Requests from https://api.mistral.ai/v1/chat/completions: Rate limit exceeded
No metrics are reported for this provider in this edition. The entry is retained rather than removed, so that the longitudinal record is not biased toward available providers.
Mistral AI
mistral-small-2603
rate limit
429 Too Many Requests from https://api.mistral.ai/v1/chat/completions: Rate limit exceeded
No metrics are reported for this provider in this edition. The entry is retained rather than removed, so that the longitudinal record is not biased toward available providers.
DeepSeek
deepseek-v4-pro
No
| Speed | 85% |
|---|---|
| First-token responsiveness | 84% |
| Token economy | 4% |
decisiveness 100%, one-word compliance 100% for every model, so not on the radar.
Picks an answer and returns it promptly. Conviction and speed.
