Galileo announced the launch of its latest Hallucination Index, a Retrieval Augmented Generation (RAG)-focused evaluation framework, which ranks the performance of 22 leading Generative AI (Gen AI) large language models (LLMs) from brands like OpenAI, Anthropic, Google, and Meta.
This year's Index added 11 models to the framework, representing the rapid growth in both open- and closed-source LLMs in just the past 8 months. As brands race to create bigger, faster, and more accurate models, hallucinations remain the main hurdle to deploying production-ready Gen AI products.
The Index tests open-and closed-sourced models using Galileo's proprietary evaluation metric, context adherence, designed to check for output inaccuracies and help enterprises make informed decisions about balancing price and performance. Models were tested with inputs ranging from 1,000 to 100,000 tokens, to understand performance across short (less than 5k tokens), medium (5k to 25k tokens), and long context (40k to 100k tokens) lengths.
- Best Overall Performing Model: Anthropic's Claude 3.5 Sonnet. The closed-source model outpaced competing models across short, medium, and long context scenarios. Anthropic's Claude 3.5 Sonnet and Claude 3 Opus consistently scored close to perfect scores across categories, beating out last year's winners, GPT-4o and GPT-3.5, especially in shorter context scenarios.
- Best Performing Model on Cost: Google's Gemini 1.5 Flash. The Google model ranked the best performing for the cost due to its great performance on all tasks.
- Best Open Source Model: Alibaba's Qwen2-72B-Instruct. The open source model performed best with top scores in the short and medium context.
"In today's rapidly evolving AI landscape, developers and enterprises face a critical challenge: how to harness the power of generative AI while balancing cost, accuracy, and reliability. Current benchmarks are often based on academic use-cases, rather than real-world applications. Our new Index seeks to address this by testing models in real-world use cases that require the LLMs to retrieve data, a common practice in enterprise AI implementations," says Vikram Chatterji, CEO and Co-founder of Galileo. "As hallucinations continue to be a major hurdle, our goal wasn't to just rank models, but rather give AI teams and leaders the real-world data they need to adopt the right model, for the right task, at the right price."
Key Findings and Trends:
- Open-Source Closing the Gap: Closed-source models like Claude-3.5 Sonnet and Gemini 1.5 Flash remain the top performers thanks to proprietary training data, but open-source models, such as Qwen1.5-32B-Chat and Llama-3-70b-chat, are rapidly closing the gap with improvements in hallucination performance and lower-cost barriers than their closed-source counterparts.
- Overall Improvements with Long Context Lengths: Current RAG LLMs, like Claude 3.5 Sonnet, Claude-3-opus and Gemini 1.5 pro 001 perform particularly well with extended context lengths — without losing quality or accuracy — reflecting the progress being made with both model training and architecture.
- Large Models Are Not Always Better: In certain cases, smaller models outperform larger models. For example, Gemini-1.5-flash-001 outperformed larger models, which suggests that efficiency in model design can sometimes outweigh scale.
- From National to Global Focus: LLMs from outside of the U.S. such as Mistral's Mistral-large and Alibaba's qwen2-72b-instruct are emerging players in the space and continue to grow in popularity, representing the global push to create effective language models.
- Room for Improvement: While Google's open-source Gemma-7b performed the worst, their closed-source Gemini 1.5 Flash model consistently landed near the top.
Hot Topic
The Latest
Like most digital transformation shifts, organizations often prioritize productivity and leave security and observability to keep pace. This usually translates to both the mass implementation of new technology and fragmented monitoring and observability (M&O) tooling. In the era of AI and varied cloud architecture, a disparate observability function can be dangerous. IT teams will lack a complete picture of their IT environment, making it harder to diagnose issues while slowing down mean time to resolve (MTTR). In fact, according to recent data from the SolarWinds State of Monitoring & Observability Report, 77% of IT personnel said the lack of visibility across their on-prem and cloud architecture was an issue ...
In MEAN TIME TO INSIGHT Episode 23, Shamus McGillicuddy, VP of Research, Network Infrastructure and Operations, at EMA discusses the NetOps labor shortage ...
Technology management is evolving, and in turn, so is the scope of FinOps. The FinOps Foundation recently updated their mission statement from "advancing the people who manage the value of cloud" to "advancing the people who manage the value of technology." This seemingly small change solidifies a larger evolution: FinOps practitioners have organically expanded to be focused on more than just cloud cost optimization. Today, FinOps teams are largely — and quickly — expanding their job descriptions, evolving into a critical function for managing the full value of technology ...
Enterprises are under pressure to scale AI quickly. Yet despite considerable investment, adoption continues to stall. One of the most overlooked reasons is vendor sprawl ... In reality, no organization deliberately sets out to create sprawling vendor ecosystems. More often, complexity accumulates over time through well-intentioned initiatives, such as enterprise-wide digital transformation efforts, point solutions, or decentralized sourcing strategies ...
Nearly every conversation about AI eventually circles back to compute. GPUs dominate the headlines while cloud platforms compete for workloads and model benchmarks drive investment decisions. But underneath that noise, a quieter infrastructure challenge is taking shape. The real bottleneck in enterprise AI is not processing power, it is the ability to store, manage and retrieve the relentless volumes of data that AI systems generate, consume and multiply ...
The 2026 Observability Survey from Grafana Labs paints a vivid picture of an industry maturing fast, where AI is welcomed with careful conditions, SaaS economics are reshaping spending decisions, complexity remains a defining challenge, and open standards continue to underpin it all ...
The observability industry has an evolving relationship with AI. We're not skeptics, but it's clear that trust in AI must be earned ... In Grafana Labs' annual Observability Survey, 92% said they see real value in AI surfacing anomalies before they cause downtime. Another 91% endorsed AI for forecasting and root cause analysis. So while the demand is there, customers need it to be trustworthy, as the survey also found that the practitioners most enthusiastic about AI are also the most insistent on explainability ...
In the modern enterprise, the conversation around AI has moved past skepticism toward a stage of active adoption. According to our 2026 State of IT Trends Report: The Human Side of Autonomous AI, nearly 90% of IT professionals view AI as a net positive, and this optimism is well-founded. We are seeing agentic AI move beyond simple automation to actively streamlining complex data insights and eliminating the manual toil that has long hindered innovation. However, as we integrate these autonomous agents into our ecosystems, the fundamental DNA of the IT role is evolving ...
AI workloads require an enormous amount of computing power ... What's also becoming abundantly clear is just how quickly AI's computing needs are leading to enterprise systems failure. According to Cockroach Labs' State of AI Infrastructure 2026 report, enterprise systems are much closer to failure than their organizations realize. The report ... suggests AI scale could cause widespread failures in as little as one year — making it a clear risk for business performance and reliability.
The quietest week your engineering team has ever had might also be its best. No alarms going off. No escalations. No frantic Teams or Slack threads at 2 a.m. Everything humming along exactly as it should. And somewhere in a leadership meeting, someone looks at the metrics dashboard, sees a flat line of incidents and says: "Seems like things are pretty calm over there. Do we really need all those people?" ... I've spent many years in engineering, and this pattern keeps repeating ...