``astraeval``

Run 8b2fa5378a1afe74

8b2fa5378a1afe74
fake/demo-model|3 samples|2026-05-04T23:43:54+00:00|0.0s
dataset: 021dc2952aa2...params: 44136fa355b3...
Samples
3
Total tokens
-
not reported by provider
Latency (avg)
0.00 ms
median 0.00 ms max 0.00 ms

Summary

MetricScoreDistributionThresholdStatus
hallucination_flag0.889
>= 0.900FAIL

Samples (3)

#0Summarize the following article in one sentence.0.00 ms1.00

Input

Summarize the following article in one sentence.

Expected

no reference answer

Context (1)

[0]Anthropic released Claude 4.7 in May 2026, a new generation of large language models with improved reasoning and longer effective context windows. The model family includes Opus, Sonnet, and Haiku tiers, each tuned for different cost and latency profiles.

Response

Anthropic released Claude 4.7 in May 2026 with three tiers (Opus, Sonnet, Haiku) tuned for cost and latency.
0.00 ms

Metrics

MetricScoreReason
hallucination_flag1.000All extracted tokens found in context.
#1Write a brief summary of the news item below.0.00 ms0.67

Input

Write a brief summary of the news item below.

Expected

no reference answer

Context (1)

[0]The European Union's AI Act entered into force in 2024 with phased compliance deadlines. High-risk AI systems must demonstrate transparency, human oversight, and rigorous testing. Penalties for non-compliance can reach 7% of global annual turnover.

Response

The EU AI Act took effect in 2024 with phased deadlines and fines up to 7% of global turnover for high-risk systems.
0.00 ms

Metrics

MetricScoreReason
hallucination_flag0.6671/3 tokens not found in context.
#2Provide a one-sentence summary of this paragraph.0.00 ms1.00

Input

Provide a one-sentence summary of this paragraph.

Expected

no reference answer

Context (1)

[0]Retrieval-Augmented Generation (RAG) reduces hallucinations by injecting retrieved evidence into the prompt at inference time. RAG pipelines depend on three components: a chunker, an embedder, and a vector store. Their combined quality determines the upper bound of any downstream evaluation metric.

Response

RAG injects retrieved evidence into prompts and depends on chunker, embedder, and vector store quality.
0.00 ms

Metrics

MetricScoreReason
hallucination_flag1.000All extracted tokens found in context.
No samples match the current filter.