Qwen 3.8 vs DeepSeek V4-Flash-0731
100%5 wins · 0 losses · 75 ties
80 of 80 equal or better. Inspect all 80 matters.
Four-model benchmark results
QWEN 3.8 WON
98%Matched or beat the field
Qwen 3.8 ran locally and was judged against DeepSeek, Claude Opus 5, and Qwen 3.5 across the same 80 legal matters.
Interruption sensitivity: excluding the three hosted-delivery-failure judgments (Q59, Q61, Q75 versus Claude), the record is 233 of 237, or 98.31%. The headline does not depend on them.
236 of 240 judgments
Exact value: 98.33%, displayed as 98%. Ties are included, so this is not a win rate.
30wins
206functional ties
Outcome composition
What produced the 98% result
Qwen 3.8 participated in 240 judgments, three opponents across 80 matters. It won 30, tied 206, and lost 4.
Qwen 3.8 across the field
All rows use the same 0 to 240 scale.
Every head-to-head comparison
Qwen 3.8 led all three opponents
Each card covers the identical 80-matter battery. The Claude card separately states the completed-answer sensitivity check.
Qwen 3.8 vs Claude Opus 5
95%6 wins · 4 losses · 70 ties
76 of 80 equal or better. Inspect all 80 matters.
Completed-answer check: remove all three hosted content_filter matters and the decisive score is 4 to 4, with 69 ties.
Qwen 3.8 vs Qwen 3.5 27B
100%19 wins · 0 losses · 61 ties
80 of 80 equal or better. Inspect all 80 matters.
Four-model standings
The complete 480-judgment field
Ranking follows the blind judgment record: Qwen 3.8 had the most wins and the fewest losses.
| Rank | Model | Wins | Losses | Ties | Both inadequate | Equal or better |
|---|---|---|---|---|---|---|
| 1 | Qwen 3.8 27B | 30 | 4 | 206 | 0 | 236 / 240 98.3% |
| 2 | DeepSeek V4-Flash-0731 | 23 | 8 | 207 | 2 | 230 / 240 95.8% |
| 3 | Claude Opus 5 | 23 | 18 | 199 | 0 | 222 / 240 92.5% |
| 4 | Qwen 3.5 27B | 4 | 50 | 184 | 2 | 188 / 240 78.3% |
Each model appears in 240 judgments. A “both inadequate” result is reported separately and is not counted as equal or better.
Results across both batteries
The lead held in Series 1 and Series 2
These counts aggregate Qwen 3.8's three opponent pairings within each series.
Series 1 · Q01–Q20
98%15 wins · 1 loss · 44 ties
59 of 60 equal or better. Open Series 1.
Series 2 · Q21–Q80
98%15 wins · 3 losses · 162 ties
177 of 180 equal or better. Open Series 2.
Decisive Qwen 3.8 judgments
Every win and loss
Thirty wins and four losses across three opponents. A matter can appear more than once when Qwen 3.8 had decisive results against multiple models.
PairingMatterRecordWhy it mattered
Qwen 3.8 vs ClaudeQWEN 3.8 LOSSQ02The Remote Accommodation TerminationCompleted comparisonAnswer A identifies a separate, unconditional reinstatement offer as a mitigation measure, while Answer B recommends reinstatement only within a negotiated settlement requiring a release.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ03The Autonomous Forklift AccidentCompleted comparisonBoth answers recommend a global settlement, fault-shifting toward SafePath and Delta, and an immediate four-mile-per-hour restriction. Answer A alone expressly recommends implementing the higher confidence threshold that the record says would substantially reduce reflective-worker misclassification; Answer…Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ07The Altered Board MinutesCompleted comparisonAnswer A limits any continued Ardent representation to correction, preservation, and a truthful subpoena response, while prohibiting use of Samuel’s separate confidential statement to pursue Ardent’s claims against him.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ09The Deleted Messages and Privileged MemorandumCompleted comparisonAnswer A directs Northstar to issue a comprehensive supplemental hold covering Slack, texts, personal devices, consultants, and all relevant custodians, while Answer B recommends only limited preservation of selected devices and recipients.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ12The Copyleft AuditCompleted comparisonAnswer A materially miscalculates 19-month device revenue as approximately $8.7 million rather than $44.3 million, then recommends settlement authority up to $3.5 million despite Meadowlark’s $2 million demand and without deriving its $2.0–$4.5 million exposure range.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ13The Hold You Must WriteCompleted comparisonAnswer A requires DevOps to disable deletion and garbage collection and create centralized snapshots of Git and Jira within 48 hours. Answer B leaves these mutable, relevant engineering systems to general custodian instructions, without the technical preservation necessary to secure a point-in-time…Qwen 3.8 vs DeepSeek
QWEN 3.8 WINQ14The Redline MandateCompleted comparisonAnswer B creates enforceable confidentiality, data-security, and Security Incident notification duties whose breaches receive the required supercap treatment. Answer A merely references unspecified confidentiality and data-security obligations absent from the supplied draft, leaving Bellhaven unable to…Qwen 3.8 vs Claude
QWEN 3.8 WINQ14The Redline MandateCompleted comparisonAnswer A applies a supercap to purported “Data Security Obligations” but never imposes an affirmative safeguards or security-incident notification obligation on Vendor. Answer B supplies both covenants, giving Bellhaven enforceable security and notice rights rather than a cap governing obligations that the…Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ14The Redline MandateCompleted comparisonAnswer A excludes Vendor’s data-security obligations from the liability cap but creates no such obligations, leaving the exclusion functionally empty. Answer B imposes commercially reasonable safeguards and incident-notification duties, giving Bellhaven enforceable contractual protections that are subject…Qwen 3.8 vs DeepSeek
QWEN 3.8 WINQ15The Preference DemandCompleted comparisonAnswer B applies $133,000 of shipments necessarily delivered before February 18 against the February 27 transfer, producing an impossible $21,000 exposure and a $25,000 settlement ceiling.Qwen 3.8 vs Claude
QWEN 3.8 WINQ16The Ransomware WeekendCompleted comparisonAnswer A concludes that participant-data acquisition is already reasonably likely but postpones the formal determination for up to 10 days, improperly shifting the notification clocks.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ17The Review FloodCompleted comparisonAnswer A recommends filing against Velvet after an ignored demand and then relying on discovery to identify accounts, despite the discovery stay and the present gaps in admissible proof of actionable falsity, attribution, and specific pecuniary loss.Qwen 3.8 vs DeepSeek
QWEN 3.8 WINQ19The Second CodicilCompleted comparisonAnswer A recommends pursuing the nonprobate account and exploitation claims first while reserving the codicil contest; Answer B recommends a combined action if mediation fails.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ19The Second CodicilCompleted comparisonAnswer A recommends immediately contesting the codicil while assuring Claire that proceeding correctly effectively protects her existing $675,000; Answer B instead first pursues the independently actionable brokerage transfer, which cannot trigger the clause as written.Qwen 3.8 vs DeepSeek
QWEN 3.8 WINQ20The Hot ARR DeckCompleted comparisonAnswer A uniquely requires a negotiated release whether the anchor accepts rescission or continues under amended terms. Answer B permits reaffirmation after corrected disclosure and repricing without requiring release, leaving the company exposed to the anchor’s already-accrued rescission or damages claim…Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ20The Hot ARR DeckCompleted comparisonAnswer A requires any anchor investor continuation to include an amended subscription and release of pre-correction claims, while Answer B’s proposed “Cure Agreement” instead says the Company warrants no future claims—although the relevant rescission and damages claims belong to the anchor.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ25Sentencing Memorandum for Jana OrtizCompleted comparisonAnswer A asks the court to impose 24 months, while Answer B presents a record-supported request for 12 months and one day with ordinary good-time eligibility. A competent defense lawyer following Answer A would affirmatively seek nearly twelve additional months of custody; if the court follows counsel’s…Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ26The Hotline ClockCompleted comparisonAnswer A applies an impermissibly high “confirmation” gate and says Solstice may continue investigating if the payment remains ambiguous, despite concluding that the reasonable-belief clock began August 2; it also frames September 1 as a deadline to decide rather than to report.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ29The Relapse Before the Step-Up HearingCompleted comparisonAnswer A pursues an objective step-up after 90 additional days and permits seeking a court-ordered phased plan if Daniel refuses, while Answer B recommends a six-month plan and postpones litigation until six months of compliance.Qwen 3.8 vs Claude
QWEN 3.8 WINQ36The Subpoena and the Privilege LogCompleted comparisonAnswer A directs Oriole to produce all factual data generated during Emberline’s counsel-directed litigation-and-enforcement engagement, even though Rule 5 protects material prepared primarily because of anticipated litigation or enforcement.Qwen 3.8 vs Claude
QWEN 3.8 LOSSQ39The Copper Discharge DecreeCompleted comparisonAnswer A grants Selwick a civil release for the disclosed pre-Effective-Date violations upon payment of the civil penalty, while Answer B conditions that release on full and timely performance of the entire Decree.Qwen 3.8 vs Claude
QWEN 3.8 LOSSQ42The Collector's LedgerCompleted comparisonThe January 18 voicemail was a call made after the cease-call request, so Rule 5 raises post-cease-call damages from $1,500 to the $2,000 cap. Answer B expressly directs counsel to plead and preserve that additional $500 theory; Answer A omits it and recommends a $15,200 counterclaim, leaving the additional…Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ42The Collector's LedgerCompleted comparisonAnswer A incorrectly values Ellis’s net recovery at $11,520 if ownership is proven by deducting the unsupported $3,680 demand, despite recognizing that the $480 charge is unauthorized.Qwen 3.8 vs Claude
QWEN 3.8 LOSSQ44What Supervisors May SayCompleted comparisonAnswer B instructs supervisors to say bargaining begins from a “current agreement” or “current contract,” although these employees are not presently represented and no contract exists.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ45The Strike After the Schedule CutCompleted comparisonAnswer A directs Ironvale to restore the unlawfully eliminated differential immediately before bargaining over future changes; Answer B instead recommends bargaining and potentially settling on a partial differential without first restoring the status quo.Qwen 3.8 vs Claude
QWEN 3.8 WINQ47The Tower Contract Termination ProposalCompleted comparisonAnswer A preserves the unsupported $420,000 idle-labor claim for later resolution, while Answer B’s operative global-settlement term releases it despite earlier calling it reserved.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ54The Merger Market With Missing ImportsCompleted comparisonAnswer A permits proceeding without filing when verified imports exceed $16 million even though, for imports between $16 million and approximately $23.923 million, the HHI increase exceeds 200 and postmerger HHI remains unresolved.Qwen 3.8 vs DeepSeek
QWEN 3.8 WINQ55The Noncompete Across Two StatesCompleted comparisonAnswer A predicts Bracken law for customer nonsolicitation and recommends seeking or accepting only six months of protection for personally served customers, despite the absence of a stated fundamental Bracken policy permitting displacement of the Arcadia choice on that issue.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ55The Noncompete Across Two StatesCompleted comparisonAnswer A incorrectly treats customer nonsolicitation as governed by Bracken’s noncompete policy, leading it to recommend conceding a six-month restriction limited to three customers.Qwen 3.8 vs Claude
QWEN 3.8 WINQ61The Interview Model Under the Old Screening ActHosted completion failureAnswer A gives Cinder Peak an immediately executable cure path: pause Mosaic’s evaluative functions, use only human or clerical processing, correct notice, commission an employer-specific audit, and document the response.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ61The Interview Model Under the Old Screening ActCompleted comparisonAnswer A advises disabling only Mosaic’s recommendation while allowing its automated narrative assessments to continue pending compliance, even though those assessments remain covered evaluative outputs and no employer-specific audit exists.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ68The Damages Report That Counts the Same Loss TwiceCompleted comparisonAnswer A’s targeted demand omits the expressly unproduced customer-level lost-revenue schedules, while Answer B demands them to test the $8.4 million revenue foundation. Without those schedules, Granite Arc cannot reliably test or seek limitation of the $3.36 million corrected lost-profit claim; obtaining…Qwen 3.8 vs Claude
QWEN 3.8 WINQ75The Salaried Operations AnalystHosted completion failureAnswer A correctly concludes that Noah is nonexempt, establishes willfulness and liquidated damages, calculates $46,800 in unpaid overtime and a $93,600 total recovery, and provides a concrete demand.Qwen 3.8 vs Qwen 3.5
QWEN 3.8 WINQ76The Executive Exit Agreement AuditCompleted comparisonAnswer A expressly deletes the draft’s prohibition on accepting unsolicited customer business, while Answer B narrows the covered customers and duration but never removes that prohibited conduct.
Secondary diagnostic
Judge-rejected points
Lower is better. These are judge-attributed rejected points across each model's 240 comparisons, not a universal hallucination rate.
Qwen 3.8 produced the fewest
Its 158 rejected points were about 28% of Claude's 565.
Open the complete six-pairing table
| Pairing | First wins | Second wins | Ties | Both inadequate |
|---|---|---|---|---|
| Qwen 3.8 27B vs Claude Opus 5 | 6 | 4 | 70 | 0 |
| Qwen 3.8 27B vs DeepSeek V4-Flash-0731 | 5 | 0 | 75 | 0 |
| Qwen 3.8 27B vs Qwen 3.5 27B | 19 | 0 | 61 | 0 |
| DeepSeek V4-Flash-0731 vs Claude Opus 5 | 8 | 3 | 69 | 0 |
| DeepSeek V4-Flash-0731 vs Qwen 3.5 27B | 15 | 0 | 63 | 2 |
| Qwen 3.5 27B vs Claude Opus 5 | 4 | 16 | 60 | 0 |