AI Agent StoreAgent Fit
A score based on LiveBench test results, giving more importance to the skills your chosen task needs. Use it to build a shortlist. It does not predict how often your complete agent will succeed.
A FIELD GUIDE TO THE NUMBERS
Less benchmark jargon.
More useful answers.
A benchmark asks a model to do a particular kind of work. Knowing what the test asks is more useful than knowing who came first.
Data reviewed · LiveBench question set: 25 June 2026 · View every source ↗ · Download the data ↓
Practical Value
Combines task ability and estimated cost: 70% Agent Fit and 30% affordability by default. We compare costs across these 61 models using your usage. Change the slider to choose how much price matters. This rating is not a savings estimate.
Evidence Consensus
Shows how the same 10 model versions and settings rank in both LiveBench tests and Arena user votes, giving each source equal weight. A blank means a matching result is missing. This is a comparison of that group, not every available model.
LiveBench
Automatically checked tasks in seven areas: reasoning, coding, agentic coding, mathematics, data analysis, language and instruction following. Overall is the equal-weight category mean. All rows use the 2026-06-25 question set.
Arena preference
People compare answers without knowing which model wrote them. A higher rating means more preferred answers, not a percent correct. Published ± ranges can overlap; small score gaps may not mean a real difference.
Reasoning
Can it work through constraints and connect clues? LiveBench uses spatial, navigation, perspective-taking and logic-puzzle tasks. Useful for planning, but not a direct test of long-running agents.
Coding
Can it write or complete code that passes tests? These are contained programming problems, not entire software projects.
Agentic coding
Can it edit code in a tool-using workflow? LiveBench tests JavaScript, TypeScript and Python tasks. Results depend on the testing software and tools as well as the model.
Mathematics
Can it solve difficult quantitative problems with checkable answers? Strong math is useful evidence of reasoning, but does not guarantee better writing or tool use.
Data analysis
Can it join and reformat tables and reason about event sequences? Useful for agents that process structured business information.
Language
Can it interpret word relationships, reconstruct plots and correct typos? This is a narrow language test, not a full measure of writing quality or multilingual coverage.
Instruction following
Can it follow requested constraints while rewriting, simplifying, summarizing and composing text? Relevant to agents that must return a specific format.
Context window
How much information a model can hold in one request, usually including your input and its response. It is measured in tokens, the pieces of text a model processes. Check your service’s exact limit. A larger window does not guarantee better understanding.
Workload cost
Estimated text API bill: requests × (input tokens × input price + output tokens × output price) ÷ 1,000,000. Include billed reasoning in output. Recorded model-specific context tiers apply; caching, tools, retries, taxes, hosting and other premiums are excluded.
Tool calling
The provider says the model can ask your application to use a tool, such as searching a database or running code. Test whether it chooses the right tool and supplies the right information.
BFCL V4
Berkeley tests selecting functions, supplying arguments, multi-turn tool use, search and memory. Its older model configurations are shown separately and never substituted for newer releases.
A DIFFERENT TEST. A DIFFERENT SNAPSHOT.
Tool-use evidence: Berkeley BFCL V4
These published configurations are from an older, separate evaluation, last updated 12 April 2026. They are useful context for understanding tool use; they do not measure the latest versions in our model table and are excluded from our aggregate ratings. FC means the model uses a built-in feature for requesting tools; Prompt means the tools are described in the instructions.
| Tested configuration | Overall accuracy | Multi-turn | Memory |
|---|---|---|---|
| Claude-Opus-4-5-20251101 (FC) | 77.47% | 68.38% | 73.76% |
| Claude-Sonnet-4-5-20250929 (FC) | 73.24% | 61.37% | 64.95% |
| Gemini-3-Pro-Preview (Prompt) | 72.51% | 60.75% | 61.72% |
| GLM-4.6 (FC thinking) | 72.38% | 68.00% | 55.70% |
| Grok-4-1-fast-reasoning (FC) | 69.57% | 58.87% | 53.98% |
| Claude-Haiku-4-5-20251001 (FC) | 68.70% | 53.62% | 54.41% |
| Gemini-3-Pro-Preview (FC) | 68.14% | 63.12% | 54.84% |
| o3-2025-04-16 (Prompt) | 63.05% | 62.25% | 51.83% |
| Grok-4-0709 (Prompt) | 62.97% | 47.00% | 50.54% |
| Grok-4-0709 (FC) | 61.38% | 33.88% | 55.91% |
GROUNDED ANSWERS NEED THEIR OWN TEST
Hallucination is not one universal number.
Vectara’s document-summary evaluation checks whether an answer stays supported by the supplied document. The 2026-09-22 snapshot below uses Vectara’s exact named configurations, not the reasoning configurations in our LiveBench table. GPT-6.1 Sol and Claude Sonnet 5.5 do not have verified matching observations here.
| Exact tested model | Hallucination rate | Factual consistency | Answer rate |
|---|---|---|---|
openai/gpt-5.4-nano-2026-03-17 | 3.1% | 96.9% | 100.0% |
openai/gpt-5.4-mini-2026-03-17 | 5.5% | 94.5% | 100.0% |
openai/gpt-6-sol | 6.5% | 93.5% | 100.0% |
openai/gpt-5.4-2026-03-05 | 7.0% | 93.0% | 99.9% |
openai/gpt-6-astra | 8.7% | 91.3% | 100.0% |
openai/gpt-5.5 | 9.3% | 90.7% | 100.0% |
These rates do not predict factual accuracy on open-ended questions or autonomous tasks. Also check how often the model answered and how the answers were graded. We therefore keep them out of our scores. Read the dataset, configurations and limitations ↗
Now put the evidence to work.
Compare models with explanations beside every metric.