> ## Documentation Index
> Fetch the complete documentation index at: https://narev.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Blog - Index

> Read the latest Narev research, product updates, and analysis on LLM cost optimization, AI FinOps, and the future of AI unit economics.

Welcome to the Narev blog. Find the latest research, product updates, and thoughts on the future of AI economics here.

<Update
  label="Cost optimization"
  description="2026-06-13"
  rss={{
title: "How we cut Dagster Cloud Serverless credits by collapsing dbt assets",
description:
  "Learn how Narev reduced its Dagster Cloud Serverless bill after the May 2026 pricing changes by collapsing its dbt project into a single asset."
}}
>
  ## [How we cut Dagster Cloud Serverless credits by collapsing dbt assets](/docs/blog/lower-dagster-bill)

  <Frame>
    <img src="https://mintcdn.com/narev/AxZvs5bDC8TZtbhn/images/blog/dagster-credits/credits-usage-annotated-well.png?fit=max&auto=format&n=AxZvs5bDC8TZtbhn&q=85&s=77a3c6d6dc440e664766b9a5c14495eb" alt="Dagster Cloud Serverless credits usage chart showing a steep drop in daily credits consumed after collapsing dbt assets" width="394" height="464" data-path="images/blog/dagster-credits/credits-usage-annotated-well.png" />
  </Frame>

  Learn how Narev reduced its Dagster Cloud Serverless bill after the May 2026 pricing changes by collapsing its dbt project into a single asset.

  [Read the full post](/docs/blog/lower-dagster-bill)
</Update>

<Update
  label="Model pricing"
  description="2025-10-15"
  rss={{
title: "Claude Fable 5 Released: 10% Better, 20x More Expensive",
description:
  "Claude Fable 5 is here. It's 10% better than Composer 2.5, but costs 20x more. See the charts and analysis behind the new model's price."
}}
>
  ## [Claude Fable 5 Released: 10% Better, 20x More Expensive](/docs/blog/claude-fable-released)

  <Frame>
    <img src="https://mintcdn.com/narev/RkLV_LSwa7JjBj8u/images/blog/claude-fable/claude-fable-annotated.png?fit=max&auto=format&n=RkLV_LSwa7JjBj8u&q=85&s=b4a1c251de89fe30bec7a7dd1070a841" alt="Annotated chart emphasizing the 10 percent performance improvement of Claude Fable 5 despite its 20x higher cost compared to Composer 2.5" width="1328" height="950" data-path="images/blog/claude-fable/claude-fable-annotated.png" />
  </Frame>

  Claude Fable 5 is here. It's 10% better than Composer 2.5, but costs 20x more. See the charts and analysis behind the new model's price.

  [Read the full post](/docs/blog/claude-fable-released)
</Update>

<Update
  label="Latest research"
  description="2025-10-14"
  rss={{
title: "GPT-3.5 MMLU Score: why it beats GPT-5 at 3% of the cost",
description:
  "Despite its low MMLU score, GPT-3.5 outperformed top 2025 models like GPT-5 and Claude Opus on a real task, costing $823 vs $30,390 per 1M requests."
}}
>
  ## [GPT-3.5 MMLU Score: why it beats GPT-5 at 3% of the cost](/docs/blog/gpt35-beats-gpt5)

  <Frame>
    <img src="https://mintcdn.com/narev/fpyXaBUjJEyahZQ-/images/blog/gpt35-beats-gpt5/hero.webp?fit=max&auto=format&n=fpyXaBUjJEyahZQ-&q=85&s=4c3f5a87dc8c616d3b2160d18ac305f6" alt="GPT-3.5 benchmark hero image" width="1508" height="924" data-path="images/blog/gpt35-beats-gpt5/hero.webp" />
  </Frame>

  Despite its low MMLU score, GPT-3.5 outperformed top 2025 models like GPT-5 and Claude Opus on a real task, costing $823 vs $30,390 per 1M requests.

  [Read the full post](/docs/blog/gpt35-beats-gpt5)
</Update>

<Update
  label="Benchmarking"
  description="2025-10-13"
  rss={{
title: "GPT-4o vs Claude Opus - MMLU Scores vs. Actual API Costs",
description:
  "MMLU scores for GPT-4o vary by 13 points, while top models differ by 1%. The real difference? Massive cost disparities for the same performance."
}}
>
  ## [GPT-4o vs Claude Opus - MMLU Scores vs. Actual API Costs](/docs/blog/mmlu-doesnt-matter)

  <Frame>
    <img src="https://mintcdn.com/narev/fpyXaBUjJEyahZQ-/images/blog/blog-post-4.webp?fit=max&auto=format&n=fpyXaBUjJEyahZQ-&q=85&s=59401fd56cbbdeb3e54d3f67529f1ddc" alt="MMLU scores versus API costs blog image" width="2400" height="1350" data-path="images/blog/blog-post-4.webp" />
  </Frame>

  MMLU scores for GPT-4o vary by 13 points, while top models differ by 1%. The real difference? Massive cost disparities for the same performance.

  [Read the full post](/docs/blog/mmlu-doesnt-matter)
</Update>

<Update
  label="Research"
  description="2025-10-11"
  rss={{
title: "AI adoption rate for large firms continues to trend down",
description:
  "U.S. Census Bureau data shows AI adoption among large firms has continued to decline after peaking in July 2025, while smallest firms keep growing"
}}
>
  ## [AI adoption rate for large firms continues to trend down](/docs/blog/ai-adoption-rate)

  <Frame>
    <img src="https://mintcdn.com/narev/fpyXaBUjJEyahZQ-/images/blog/ai-adoption-rate/ai-adoption-rate-chart.webp?fit=max&auto=format&n=fpyXaBUjJEyahZQ-&q=85&s=d2ef2f698c90286ea706c28364b36da0" alt="AI adoption rate chart for large firms" width="1874" height="1465" data-path="images/blog/ai-adoption-rate/ai-adoption-rate-chart.webp" />
  </Frame>

  U.S. Census Bureau data shows AI adoption among large firms has continued to decline after peaking in July 2025, while smallest firms keep growing.

  [Read the full post](/docs/blog/ai-adoption-rate)
</Update>

<Update
  label="Company"
  description="2025-07-22"
  rss={{
title: "Why Narev AI Exists - Two futures for AI economics",
description:
  "There's a world where AI gets infinitely cheaper, and another where users need to choose the best tool for every job. Narev is building for the latter."
}}
>
  ## [Why Narev AI Exists - Two futures for AI economics](/docs/blog/why-we-launch)

  <Frame>
    <img src="https://mintcdn.com/narev/fpyXaBUjJEyahZQ-/images/blog/blog-post-3.webp?fit=max&auto=format&n=fpyXaBUjJEyahZQ-&q=85&s=4086b73d7d1344601019743d5d39fc87" alt="Why Narev exists blog image" width="2400" height="1350" data-path="images/blog/blog-post-3.webp" />
  </Frame>

  There's a world where AI gets infinitely cheaper, and another where users need to choose the best tool for every job. Narev is building for the latter.

  [Read the full post](/docs/blog/why-we-launch)
</Update>