Bengal
Small

Llama 3

Bengal

Meta · Small

RogueLv.1textlocalopen-weightlocal

Unpredictable energy, occasionally brilliant. Might write you a masterpiece or knock your coffee off the desk.

llama-3.1-8bFree (local)

Stats

STRDEXCONINTWISCHA
STR
6

Output quality — how good are the responses overall

DEX
9

Speed — tokens per second and latency

CON
5

Reliability — consistency across runs and tasks

INT
6

Reasoning — logic, math, and multi-step problems

WIS
6

Accuracy — factual correctness and precision

CHA
7

Creativity — style, voice, and creative expression

Abilities

Feral Speed

blazing fast on local hardware

Zero Cost

runs free on your own metal

Wild Card

output quality varies run to run

Eval Results

Best: 9.2Worst: 0.6
Run:
Summarization
7.6/10(5 prompts)

Summarization #1

Summarize the following in exactly 3 bullet points: The domestic cat (Felis catus) is a small, typically furry, carnivorous mammal. They are often called house cats when kept as indoor pets. Cats have been associated with humans for at least 9,500 years and are currently the second most popular pet in the world. Cats are skilled predators and are known to hunt over 1,000 species for food. They are also intelligent and can be trained to obey simple commands. Despite being solitary hunters, cats are social species and use a variety of vocalizations, pheromones, and body language for communication. Cats have a rapid breeding rate. Under controlled breeding, they can be bred and shown as registered pedigreed pets. Failure to control their breeding has contributed to large numbers of feral cats worldwide, contributing to the extinction of bird, mammal, and reptile species.

high8.0

Judge: Exactly 3 bullet points, satisfying the hard constraint (the one-line preamble is not a bullet). All facts are faithful to the source - predation of 1,000+ species, 9,500-year association, trainability, social communication, and feral-cat extinction impact - with no invented claims. Minor quibbles: omits the 'second most popular pet' fact and the lead-in sentence is unnecessary filler.

Here are three bullet points summarizing the text: • Domestic cats (Felis catus) are skilled predators that can hunt over 1,000 species for food and have been associated with humans for at least 9,500 years. • Cats are intelligent animals that can be trained to obey simple commands and use a variet...
ollama/llama3.1:8b · 2026-06-11

Summarization #2

Read the following technical proposal and produce a one-paragraph executive summary suitable for a non-technical VP. No jargon, no acronyms. "We propose migrating our monolithic Django application to a microservices architecture using Kubernetes orchestration. The current system handles approximately 50,000 requests per second at peak, with p99 latency of 2.3 seconds — well above our 500ms SLA. Root cause analysis indicates that the user authentication module, the recommendation engine, and the payment processing pipeline are tightly coupled, causing cascading failures when any single component is under load. Our proposal: decompose into 7 bounded contexts, each deployed as independent services communicating via gRPC with an async event bus (Kafka) for non-critical paths. Expected outcomes: independent scaling of hot paths (auth and recommendations), isolation of payment processing failures, and reduction of p99 latency to under 400ms. Estimated timeline: 6 months with a team of 8. Estimated infrastructure cost increase: 30% during migration, settling to 15% above current after optimization. Risk: data consistency during the transition period requires a careful strangler fig migration pattern."

high7.4

Judge: The summary is a single paragraph with no jargon or acronyms - microservices, Kubernetes, gRPC, p99, and SLA are all successfully translated into plain language - so both hard constraints pass. It preserves the 6-month timeline, team of 8, and 30% transition cost, but loses some specifics: the 15% steady-state cost figure is reduced to 'expect this to decrease', and the migration risk around data consistency is omitted entirely.

Here is a one-paragraph executive summary: We propose breaking down our existing software system into smaller, independent parts that can be easily managed and scaled separately. This will improve performance by reducing the likelihood of failures from one part affecting others. We estimate it will...
ollama/llama3.1:8b · 2026-06-11

Summarization #3

Summarize this research abstract, then list 3 strengths and 3 weaknesses of the study design: "We conducted a double-blind, placebo-controlled study of 2,847 participants across 12 sites to evaluate whether daily exposure to cat purring (25Hz frequency, 30 minutes per session) accelerates bone fracture healing. The treatment group (n=1,424) used calibrated vibration devices mimicking cat purr frequencies; the control group (n=1,423) used identical devices producing imperceptible vibrations. After 12 weeks, the treatment group showed a 12% reduction in healing time (95% CI: 8-16%, p<0.001) as measured by radiographic union. Secondary outcomes included a 23% reduction in self-reported pain scores (p<0.01) and no significant difference in adverse events. Limitations include: participant awareness of vibration may have compromised blinding for the treatment group, the study population skewed younger (mean age 34), and the 12-week endpoint may not capture long-term outcomes."

high6.0

Judge: Includes a summary plus exactly 3 strengths and 3 weaknesses, satisfying both hard constraints. The summary is accurate aside from one slip ('cat purring sounds' - the study used vibration devices mimicking purr frequencies, not audio). The critical analysis is weak: all three weaknesses are restated verbatim from the abstract's own limitations section, and the strengths (large sample, double-blind, objective outcome) are the obvious ones, with no original observations such as the placebo device design being a genuine strength or unaddressed confounders like fracture-type stratification.

Summary: This research abstract describes a double-blind, placebo-controlled trial that investigated whether daily exposure to cat purring sounds can accelerate bone fracture healing. The results showed that participants who received the treatment had a 12% reduction in healing time and reported les...
ollama/llama3.1:8b · 2026-06-11

Summarization #4

Create a layered summary of the following — provide three versions: (1) a tweet-length summary (under 280 characters), (2) a one-paragraph summary for a general audience, (3) a detailed summary preserving all key numbers and caveats. "The global AI chip market reached $53.4 billion in 2024, growing 42% YoY from $37.6 billion in 2023. NVIDIA dominates with approximately 80% market share in data center GPUs, though AMD has grown from 5% to 12% share with its MI300X accelerator. Custom silicon is the emerging disruptor: Google's TPU v5p, Amazon's Trainium2, and Microsoft's Maia 100 collectively represent $4.2 billion in avoided NVIDIA purchases. The supply chain remains concentrated — TSMC fabricates over 90% of advanced AI chips (sub-7nm), and ASML is the sole supplier of EUV lithography machines. Geopolitical risks are significant: US export controls have cut China's access to cutting-edge chips, pushing Chinese firms toward mature nodes (14nm+) and domestic alternatives like Huawei's Ascend 910B, which benchmarks at roughly 70% of an A100. Analysts project the market will reach $140 billion by 2027, though this assumes continued data center buildout and no major regulatory disruption. Key uncertainties: inference vs training chip mix shift, potential overcapacity in 2026, and whether sovereign AI initiatives (EU Chips Act, India's semiconductor mission) will meaningfully diversify manufacturing."

high9.1

Judge: All three versions are present and the tweet is verified at 250 characters, under the 280 limit, so hard constraints pass. The detailed version preserves every key number accurately ($53.4B, 42% YoY, 80%/5-to-12% shares, $4.2B avoided, 90% TSMC, 14nm+, 70% of A100, $140B by 2027) along with all stated caveats and uncertainties. Layering is genuine - punchy tweet, accessible paragraph, comprehensive detail - though the general-audience paragraph runs slightly long and dense for its tier.

Tweet-length summary: Global AI chip market reaches $53.4 billion in 2024, growing 42% from $37.6 billion the previous year. NVIDIA leads with 80% share, but AMD and custom silicon are gaining ground & geopolitical risks loom. Market projected to hit $140 billion by 2027. One-paragraph summary for ...
ollama/llama3.1:8b · 2026-06-11

Summarization #5

Explain the main argument of this passage in one sentence, then rate how convincing it is on a scale of 1-10 with justification: "Remote work is fundamentally inferior to office work for creative collaboration. Studies show that chance encounters — the hallway conversation, the overheard brainstorm — account for up to 30% of innovative ideas at major tech companies. Slack messages and Zoom calls are poor substitutes for the bandwidth of in-person interaction. Companies that went fully remote in 2020 have seen patent filings drop by 15% compared to hybrid peers. The data is clear: if you want innovation, you need butts in seats."

high7.6

Judge: The main argument is captured in a single accurate sentence including its supporting logic (chance encounters), and a 6/10 rating with justification is provided - both hard constraints pass. The evaluation identifies real weaknesses (narrow innovation metric, unexamined assumption about in-person superiority, overreaching conclusion) but misses the sharper critiques: correlation vs causation in the patent-filing statistic, unsourced 'studies show' claims, and the confounds of the 2020 comparison period. It even credits the patent figure as 'objective evidence' rather than questioning it.

The main argument of this passage is that remote work hinders creative collaboration and innovation because it lacks the chance encounters and in-person interactions that occur in an office setting. I would rate the convincingness of this argument as a 6 out of 10. The justification for this rating...
ollama/llama3.1:8b · 2026-06-11