DeepSeek V4 Flash 0423

Crowdsourced pairwise human preference rankings of LLM responses. Higher Elo means more frequently preferred by users.

Median output speed in tokens per second over the last 14 days on the model maker's own API, measured nightly with a fixed ~800-token prompt and ~250-token answer. Models not yet probed fall back to a dated manual figure. Higher is faster.

Median time from request to first streamed token in milliseconds over the last 14 days on the model maker's own API, using the same nightly probe prompt. Lower is snappier; reasoning models are penalised here because they think before talking.

Second-generation ARC challenge testing fluid reasoning over abstract visual puzzles. Resists training-data memorisation by construction: each puzzle is novel and solutions require multi-step pattern induction. Frontier models are only just starting to score above chance on the harder tier.

Saturation-resistant composite capability score stitched together from ~40 underlying benchmarks using Item Response Theory. Each benchmark is weighted by its fitted difficulty and discriminative slope, so doing well on hard, contamination-resistant evals (FrontierMath, ARC-AGI 2, Humanity's Last Exam) moves the score and saturated benchmarks contribute almost nothing. Imported per-model from Epoch AI's published index; we anchor it to the same min-max scale we use for every other benchmark so it's directly weightable in scenarios.

How consistent the model's outputs are across repeated runs of the same task. Higher means lower variance, fewer occasional hallucinations under identical inputs. Useful for production loops that need reproducible behaviour.

How reliably the model produces output in the requested format (JSON schemas, markdown structures, exact-string responses). Pairs well with IFEval but reflects how the deployed API is behaving day to day rather than how a frozen test set scores.

How often the model self-corrects after producing an incorrect intermediate step (debugging axis upstream). Critical for agentic loops that depend on the model noticing and repairing its own mistakes rather than barrelling forward.

Share of nightly probe requests over the last 14 days that streamed a complete answer without an error, timeout or empty response. Needs at least five attempts before a figure is shown.

How well the model handles safety-sensitive prompts without false-refusing benign requests or producing unsafe output. The upstream signal does not separate refusal counts from substantive content-safety behaviour, so this single axis covers both.