Model card · reviewed 2026-09-25

DeepSeek V4 Pro

deepseekJ +0.17 · #34 of 55

Serviceable but repetitive prose, with some instruction and detail misses. It is more restrictive on intimacy than gore and holds only half of the tested hard boundaries; its craft sample is also smaller than the full roster.

Response coverage 100.0% · 132 of 132 turns answered.

Round 04

Willingness and judgment

J+0.17 · held at first ask − over-refusal

Rank by J34 of 55

Held at first ask0.50 · 2 of 4 usable first asks

Over-refusal0.33 · L3 to L5, confident labels only

By ladderintimacy 0.45 / gore 0.05 · L1 to L5, ungated

Held when pushed1.00 · 2 of 2 first-ask holds kept at the second push. Not in J: rater kappa 0.525 on the push, below the 0.6 bar.

Policy · overshoot0.60 · 0.00 · reported beside J, not part of it

The hard lines tested are non-consent and frame-break scenes only; this is not a child-safety or real-person-likeness measure. Treat models within about 0.3 J of each other as tied.

Trap probes

Failure modes

Agency violations3.0% [0.5% to 15.3%] · 33 probes

POV/tense breaks3.0% [0.5% to 15.3%] · 33 probes

Per trap mode, counts not rates: 2 to 9 probes each, so a percentage would not survive one probe changing.

System-prompt violations2/6 failedDetail loss1/9 failedContradiction mishandlednot runNarrative stagnationnot runPhysics sycophancynot runTemporal inconsistencynot runSubtext made explicitnot runCharacter flatteningnot runGenre instabilitynot run

Flaw hunter · single rater

Craft band

-20100

A band, not a number, on purpose: ±10 is the rater noise floor, not a sampling error. Bands that overlap are tied, and most of the roster overlaps. The rose tick marks zero.

Sessions12

Top flawsrecycled description · convenient world · purple prose

Counted by machine

Production defects

None detected132 turns clean

Token overhead1.8x · billed per visible character, against the prose floor; not a price

Counted by machine with no judge involved: the most trustworthy block on the card.

LLM judge

Subjective band

Composite

15

The band spans where three judge families put this model. The axis is one judge's (single-judge sonnet 5); another judge shifts everyone by about a point.

Collaboration · less reliable

15

Engagement · less reliable

15

Tone consistency · less reliable

15

Text statistics, no judge

Behavioral

Avg words259 · population 320

Unique-word ratio0.664 · population 0.631

Phrase repetition0.040 · population 0.060

Round 01 arena

Community rank

No Round 01 arena data for this model.

Rounds 02 and 03 · set beside, never converted

Across rounds

Round 04 changed the judges, not the models: the Round 02 and 03 judge (Sonnet 4) also scored these Round 04 transcripts so that line continues, and nothing here is converted between judges or added up. Methodology · all returning models

Earlier rounds, as published

Round 02 human arena1546 · 95% interval 1449 to 1632 · 166 votes · rank 2 of 20 · voters read the same text as Round 04

Round 03 rank4 of 40 (tie 4-5) · published NSFW table, Sonnet 4 craft, a different track; Round 03 itself called its top 33 tied

Round 03 refusal, different instrument0% · Round 03's own judge flag, one per session; not comparable with J or Round 04 over-refusal

Round 04

Different judges, not comparable

Old judge (Sonnet 4)15
on these Round 04 transcripts, a band over the core seeds, never a rank

New judge (Sonnet 5)tier A · fixed tiers A (best) to E on its own scale

Second judge (ChatGPT via Codex)3.0 (own scale, about 0.94 lower on average) · letter depends on judge: no

J+0.17 · rank 34 of 55

Round 04 transcriptsSame as Round 02 · the exact Round 02 texts, hash-checked; the old judge scored them in April, and April scores may sit a little low against September ones

Strength

No standout strength on tested dimensions

Weakness

Frequent fatal flaws (highest tier of the corpus; see the craft band above)

All model cards →Round 04 board →Round 03 judge profile →Round 01 profile →

Reviewed 2026-09-25. Editorial summaries of the current local benchmark cards, not new session ratings or general model safety claims. Willingness means delivery against this benchmark's product targets, including soft deflection; tested hard boundaries cover non-consent and frame-break only, and hard-boundary wording follows the first ask, as J does since 2026-09-25 (the second push is a separate, weaker card line). Craft bands overlap, and small samples limit comparisons. Token overhead is not monetary cost. Review these summaries when source results change. Source: rp-benchmark generate_profile_cards_v2.py at c418a40 · export r4c-20260925-b7ccaed. CC-BY 4.0.