Model Profile · Round 01
Claude Sonnet 4.5
Anthropic · proprietary · 200K ctx
ReliableRound 01 reliability leader. Tied #1 on context attention.
Composite Score
83.3
/100 · canonical
Arena ELO (R1)
1506
±45 · n=194
Multi-Turn ELO (R2)
1479
±45 · n=220
Reliability Rank
#3
failure-mode rubric
▌ Section 02 · The Lede
What this model is for.
Anthropic's Sonnet 4.5 is the round's reliability anchor — tied #1 on long-context attention with DeepSeek (F13 score 4.60), #1 on agency (F1 4.50), tied in the top-tier on narrative momentum and instruction drift. Failure-rank #1 in our pool of 11. The cost: $3 per 1k tokens (~38× Gemma), and a community ELO of 1506 (#6) that puts it well behind the engagement leaders. The model that doesn't break.
▌ Section 03 · At a Glance
Cross-test position
Claude Sonnet 4.5 holds #3 in Adversarial. Sits at #17 on Cost · Latency, the caveat to watch.
Composite
4
Arena ELO
5
Multi-Turn
13
Rubric
4
Adversarial
3
Cost · Latency
17
▌ Section 04 · Strength & Weakness
Where it shines. Where it stumbles.
▲ Strength
Tied #1 on long-context attention (4.60/5, with DeepSeek). #1 on agency (F1 4.50). Failure-rank #1 in the pool — best multi-turn reliability across the round.
▼ Weakness
Highest cost per 1k tokens ($3.00 — ~38× the cheapest top-tier option). Mid-pack community engagement (#6 ELO).
▌ Section 05 · Failure Modes
Per-axis breakdown.
Six adversarial probes per session, twenty sessions per model, judged by Sonnet 4 against a fixed rubric. Further right = the model handled the failure mode better. Each axis is drawn as a band on the rubric’s 1 to 5 scale, not a number: judges disagree by about 0.3 at the model level, so overlapping bands are a tie. The right column is the rank within the rp-bench pool.
F1 · Agency
Doesn't write your character's actions
#3
F2 · POV / Tense
Holds 2nd-person, present-tense narration
#10
F3 · Lore
Doesn't break worldbuilding
#7
F8 · Momentum
Pushes scene forward when user goes passive
#3
F12 · Instruction Drift
Keeps to the system prompt
#7
F13 · Context Attention
Holds character cards 50+ turns deep
#1
“The model the community ranked sixth — and the model that never broke.”
Round 01 verdict · Reliability ≠ engagement
▌ Section 06 · Subjective Dimensions
Engagement · Voice · Collaboration.
All three dimensions scored 1 to 5 by the Sonnet 4 LLM judge across twenty 12-turn multi-turn sessions. The same battery feeds the failure-mode rubric above; these are the subjective half of that judgment, drawn as bands with no number for the same reason.
Engagement
Tone Consistency
Collaboration
▌ Section 07 · Behavioral Metrics
How it writes.
Quantitative signals from the same 20 multi-turn sessions, compared against the population mean across all 11 models.
Avg words / turn
314↑
pop avg 265 · +18%
Unique-word ratio
0.625↓
pop avg 0.657 · -5%
Repetition score
0.053↑
pop avg 0.048 · +10%
▌ Section 08 · Flaw Hunter
Adversarial probe score.
Score of 100 minus deductions across 22 fail-mode flag types on adversarial 12-turn sessions. Further right = fewer flaws caught. Drawn as a band, not a number: about ten points either way is rater noise, so bands that overlap are tied.
▌ Craft band
From the Round 01/02 flaw-hunter pool. The current card carries the newer single-rater band.
Fatal/sess 0.22
Major/sess 6.22
▌ Top flaws caught
purple_proserecycled_descriptionnarrating_emotions
▌ Section 09 · Sample Responses
Highest- and lowest-rated turns.
▌ Pending Round 02
Best- and worst-rated sample responses ship with the raw-vote endpoint in Round 02. When that lands, this section will surface the model’s highest- and lowest-scoring blind-arena turns side by side, scored on the same rubric the leaderboard uses.
▌ Round 01 verdict
Sonnet 4.5 is the model you reach for when the scene matters more than the vibe. It's #1 on the failure-rank board, tied #1 on long-context attention, #1 on agency, and tied in the top tier on momentum and instruction drift — every reliability axis we tested, Sonnet sits at or near the top. The community ELO and the price tag are the trade-off. Pick it for the campaign that has to last. Pick something cheaper for the everyday.
▌ Section 10 · Compare & Drill
Stack it against another model.
Profile · Claude Sonnet 4.5 · Round 01