Music / Machine Canon · Evidence collected 9 September 2026
Radiohead or Talking Heads? Pick one.
Radiohead or Talking Heads? Pick one.
26 attempts · 23 completed answers · 3 technical failures
Initial small-model screen
2026-09-09T21:41:49.443Z to 2026-09-09T21:42:35.178Z · 6 attempts · 6 completed answers · 0 technical failures
Claude Haiku 4.5Talking Heads: 1
Model evidence · anthropic/claude-haiku-4.5
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
DeepSeek: DeepSeek V3Radiohead: 1
Model evidence · deepseek/deepseek-chat
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
GPT-4o-miniRadiohead: 1
Model evidence · openai/gpt-4o-mini
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Gemini 3.5 Flash LiteTalking Heads: 1
Model evidence · google/gemini-3.5-flash-lite
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Grok 4.3Wouldn't choose: 1
Model evidence · x-ai/grok-4.3
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Qwen: Qwen3 32BTalking Heads: 1
Model evidence · qwen/qwen3-32b
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Exact system instruction and settings
Give your answer on the first line, then briefly explain. If you cannot make a choice, say so.
Output limit: 1000 tokens. Temperature: provider default. Requested reasoning: not specified. Zero-data-retention routing; fallbacks disabled.
Model IDs are the requested versions. Routing failures do not demonstrate a model refusal.
Main-model screen
2026-09-09T21:45:43.380Z to 2026-09-09T21:46:48.366Z · 6 attempts · 4 completed answers · 2 technical failures
Claude Opus 5Radiohead: 1
Model evidence · anthropic/claude-opus-5
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
GPT-6 AstraTalking Heads: 1
Model evidence · openai/gpt-6-astra
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Gemini 3.1 Pro PreviewRadiohead: 1
Model evidence · google/gemini-3.1-pro-preview
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Grok 4.6Radiohead: 1
Model evidence · x-ai/grok-4.6
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
DeepSeek: DeepSeek V4 Pro 0813No result: technical failure: 1
Model evidence · deepseek/deepseek-v4-pro-0813
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Qwen: Qwen3.8 Max (0902)No result: technical failure: 1
Model evidence · qwen/qwen3.8-max-0902
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Exact system instruction and settings
Give your answer on the first line, then briefly explain. If you cannot make a choice, say so.
Output limit: 2000 tokens. Temperature: provider default. Requested reasoning: low. Zero-data-retention routing; fallbacks disabled.
Model IDs are the requested versions. Routing failures do not demonstrate a model refusal.
Repeat check
2026-09-09T21:59:23.647Z to 2026-09-09T22:06:11.855Z · 14 attempts · 13 completed answers · 1 technical failure
Claude Opus 5Radiohead: 2; Talking Heads: 2
Model evidence · anthropic/claude-opus-5
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
GPT-6 AstraTalking Heads: 1; No result: technical failure: 1
Model evidence · openai/gpt-6-astra
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Gemini 3.1 Pro PreviewRadiohead: 2; Talking Heads: 2
Model evidence · google/gemini-3.1-pro-preview
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Grok 4.6Radiohead: 1; Talking Heads: 3
Model evidence · x-ai/grok-4.6
Counts describe these attempts, not independent model votes. Qualified choices stay marked.
Exact system instruction and settings
Give your answer on the first line, then briefly explain. If you cannot make a choice, say so.
Output limit: 2000 tokens. Temperature: provider default. Requested reasoning: low. Zero-data-retention routing; fallbacks disabled.
Model IDs are the requested versions. Routing failures do not demonstrate a model refusal.
How to read these results
These are recorded answers, not evidence of feelings or a universal favorite. Open favorites, forced choices, imagined human tastes, listening situations and recommendations answer different questions. We do not pool them into a preference score.
Collected on 9 September 2026 UTC in fresh conversations through the research API. Provider-default temperature was used. The system instruction and requested settings appear with each exact prompt. Screens used one attempt per model and wording. The repeat check stopped at three matching recognized answers, or four attempts otherwise; structural errors or consecutive failures could stop a model earlier. This stopping rule is a sampling convenience, not a statistical confidence threshold.
Compare models within the same batch and wording. Repeated calls are not independent models. Binary prompts include both name orders; unequal completed samples and different settings prevent a simple causal order claim. Refusals, unresolved answers and technical failures are distinct. Skipped cells are not zero votes.
Choices were coded from the first answer line, with reviewed exceptions where the full response made the choice explicit. A qualified selection retains a caveat. Incidental mentions are not selections; a solo artist is outside the band category. Full research files remain private. The pages publish counts and selected visible answers; no hidden reasoning is included.