Sample data: Phase-1 sample snapshot. Official crawling and weekly benchmark jobs are not connected yet. All price, latency and score values validate the product structure only and must be replaced by traceable production data before launch.

Model comparison

The `models=a,b,c` URL parameter already drives the comparison page; selectors and saved comparisons come next.

DeepSeek V3Qwen 2.5 72BGPT-4oClaude 3.5 SonnetGemini 2.0 FlashDoubao ProKimi K2GLM-4 PlusLlama 3.1 405BMistral Large
ModelInputOutputTTFTContextValueUpdated
70Llama 3.1 70BMeta · open$0.88/1M$0.88/1M172ms128K842026-06-09Compare
DSDeepSeek V3DeepSeek · closed$0.14/1M$0.28/1M124ms128K962026-06-09Compare
70

Llama 3.1 70B

Meta · Open model served through third-party providers.

Quality
80
Chinese
69
DS

DeepSeek V3

DeepSeek · Strong value baseline for coding and Chinese tasks in the sample set.

Quality
86
Chinese
93

Turn model selection into a reviewable data workflow

Phase 1 has the interface, data fields and API skeleton; the next phase connects real crawlers, benchmark jobs and storage.

MethodologyPro