DeepSeek

The maker's own endpoint for DeepSeek V4.1 Flash. Every other provider is compared with it.

Reference

Same answer as the maker

typical for the others

DeepSeekmakersame answer twice 98% of the time

Wafer · us94%100 of 106 prompts · typical 92%

StreamLake · fp894%100 of 106 prompts · typical 92%

DeepInfra · fp893%99 of 106 prompts · typical 92%

DeepInfra · fp8 flex92%98 of 106 prompts · typical 92%

Morph · fp892%97 of 106 prompts · typical 93%

Relace · fp492%97 of 106 prompts · typical 93%

Particle AI · us90%95 of 106 prompts · typical 93%

Particle AI · us flex89%94 of 106 prompts · typical 93%

Ionstream · flex87%92 of 106 prompts · typical 93%

Every provider gets the same short prompts at temperature 0, so settings cannot change the answer. Honest providers still differ a little, which is why we compare with what is typical, not with 100%.

Question by question

DeepSeek (maker)Wafer · us StreamLake · fp8 DeepInfra · fp8 DeepInfra · fp8 flex Morph · fp8 Relace · fp4 Particle AI · us Particle AI · us flex Ionstream · flex

key94%94%93%92%92%92%90%89%87%

106 questions. Each square is one question. Lit means the same answer as DeepSeek. Hover or tap a column.

History

one cell per daily check

Listing

declared precision: unknown
price per 1M: $0.15 in · $0.6 out

Check	How	Measured	Expected	Result
Top answer	106 prompts at temperature 0. How often the top answer equals DeepSeek's own endpoint, next to how often the other providers do	reference, repeats agree 98%		pass
Answer spread	620 sampled answers compared with DeepSeek's own endpoint. Very sensitive: it also picks up harmless setup differences, so it is shown for reference only	reference	for reference	info
Word probabilities	Top token probability at temperature 0, compared with the other providers	gap 0.24	returned	pass
Pinned routing	Request pinned to this endpoint with fallbacks off	served by this provider	served by this provider	pass
Reasoning switch	Asked for no reasoning on a one word answer	respected	respected	pass

Provider reply

Reply as DeepSeek

No reply yet.

Results say whether a provider's answers match the maker's own endpoint. They are not a claim of intent. Quality checks are not running yet.