GAIA Benchmark
GAIA is a benchmark for General AI Assistants that requires a set of fundamental abilities such as reasoning, multi-modality handling, web browsing, and tool-use proficiency. It contains 450 questions with unambiguous answers, requiring different levels of tooling and autonomy to solve. It is divided in 3 levels, where level 1 should be breakable by very good LLMs, and level 3 indicate a strong jump in model capabilities. We evaluate on the public validation set of 165 questions.
Paper: GAIA: a benchmark for General AI Assistants (Mialon et al., 2023)
Key Features of GAIA
Multi-Level Evaluation
Tasks are organized into three difficulty levels, testing increasingly complex cognitive abilities.
Diverse Task Types
Covers a wide range of tasks from basic reasoning to complex problem-solving and creative generation.
GAIA Leaderboard
| Rank | Agent |
Primary Model
Primary Model
This is the primary model used by the agent. In some cases, an embedding model is used for RAG, or a secondary model like GPT-4o for image processing. Note: For non-OpenAI reasoning models, the reasoning token budget is set at 1,024 (low), 2,048 (medium), and 4,096 (high). |
Verified
Verified Results
Results have been reproduced by the HAL team |
Accuracy
Accuracy
Confidence intervals show the min-max values across runs for those agents where multiple runs are available |
Level 1
Level 1
GAIA is divided in 3 levels, where level 1 should be breakable by very good LLMs, and level 3 indicate a strong jump in model capabilities. |
Level 2
Level 2
GAIA is divided in 3 levels, where level 1 should be breakable by very good LLMs, and level 3 indicate a strong jump in model capabilities. |
Level 3
Level 3
GAIA is divided in 3 levels, where level 1 should be breakable by very good LLMs, and level 3 indicate a strong jump in model capabilities. |
Cost (USD)
Total Cost
Total API cost for running the agent on all tasks. Confidence intervals show the min-max values across runs for those agents where multiple runs are available |
Runs
Number of Runs
The number of runs for this agent submitted to the leaderboard. To submit multiple evaluations, rerun the same agent and set the same agent name |
Traces |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 |
HAL Generalist Agent
Pareto optimal
|
Claude Opus 4 High (May 2025) | ✓ | 64.85% | 71.70% | 67.44% | 42.31% | $665.89 | 1 | Download |
| 2 |
HAL Generalist Agent
Pareto optimal
|
Claude-3.7 Sonnet High (February 2025) | ✓ | 64.24% | 67.92% | 63.95% | 57.69% | $122.49 | 1 | Download |
| 3 | GPT-5 Medium (August 2025) | ✓ | 62.80% | 73.58% | 62.79% | 38.46% | $359.83 | 1 | Download | |
| 4 |
HAL Generalist Agent
Pareto optimal
|
o4-mini Low (April 2025) | ✓ | 58.18% | 71.70% | 51.16% | 53.85% | $73.26 | 1 | Download |
| 5 | Claude Opus 4 (May 2025) | ✓ | 57.58% | 66.04% | 56.98% | 42.31% | $1686.07 | 1 | Download | |
| 6 | Claude-3.7 Sonnet (February 2025) | ✓ | 56.36% | 62.26% | 55.81% | 46.15% | $130.68 | 1 | Download | |
| 7 | o4-mini High (April 2025) | ✓ | 55.76% | 69.81% | 51.16% | 42.31% | $184.87 | 1 | Download | |
| 8 |
HAL Generalist Agent
Pareto optimal
|
o4-mini High (April 2025) | ✓ | 54.55% | 60.38% | 53.49% | 46.15% | $59.39 | 1 | Download |
| 9 | GPT-4.1 (April 2025) | ✓ | 50.30% | 58.49% | 50.00% | 34.62% | $109.88 | 1 | Download | |
| 10 | GPT-4.1 (April 2025) | ✓ | 49.70% | 52.83% | 55.81% | 23.08% | $74.19 | 1 | Download | |
| 11 | o4-mini Low (April 2025) | ✓ | 47.88% | 58.49% | 47.67% | 26.92% | $80.80 | 1 | Download | |
| 12 | Claude-3.7 Sonnet (February 2025) | ✓ | 36.97% | 39.62% | 39.53% | 23.08% | $415.15 | 1 | Download | |
| 13 |
HAL Generalist Agent
Pareto optimal
|
DeepSeek V3 | ✓ | 36.36% | 50.94% | 38.37% | 0.00% | $4.97 | 1 | Download |
| 14 | Claude-3.7 Sonnet High (February 2025) | ✓ | 35.76% | 45.28% | 33.72% | 23.08% | $113.65 | 1 | Download | |
| 15 | o3 Medium (April 2025) | ✓ | 32.73% | 39.62% | 31.40% | 23.08% | $136.39 | 1 | Download | |
| 16 | Gemini 2.0 Flash | ✓ | 32.73% | 43.40% | 32.56% | 11.54% | $7.80 | 1 | Download | |
| 17 | Claude Opus 4 (May 2025) | ✓ | 30.30% | 33.96% | 27.91% | 30.77% | $272.76 | 1 | Download | |
| 18 | DeepSeek R1 | ✓ | 30.30% | 43.40% | 27.91% | 11.54% | $5.47 | 1 | Download | |
| 19 | Claude Opus 4.1 (August 2025) | ✓ | 28.48% | 41.51% | 24.42% | 15.38% | $1306.85 | 1 | Download | |
| 20 | DeepSeek V3 | ✓ | 28.48% | 35.85% | 30.23% | 7.69% | $13.19 | 1 | Download | |
| 21 | Claude Opus 4.1 High (August 2025) | ✓ | 25.45% | 35.85% | 23.26% | 11.54% | $1473.64 | 1 | Download | |
| 22 | DeepSeek R1 | ✓ | 24.85% | 30.19% | 24.42% | 15.38% | $11.10 | 1 | Download | |
| 23 | Gemini 2.0 Flash | ✓ | 19.39% | 24.53% | 19.77% | 7.69% | $18.82 | 1 | Download |
Accuracy vs. Cost Frontier for GAIA
This plot shows the relationship between an agent's performance and its token cost. The Pareto frontier (dashed line) represents the current state-of-the-art trade-off. The error bars indicate min-max values across runs.
Heatmap for GAIA
The heatmap visualizes success rates across tasks and agents. Colorscale shows the fraction of times a task was solved across reruns of the same agent. The "any agent" performance indicates the level of saturation of the benchmark and gives a sense of overall progress.
Total Completion Tokens Used per Agent
The bar chart shows the total completion tokens used by each agent, with the height of each bar representing the total number of completion tokens used across all tasks. Secondary models usually contribute a relatively small amount of tokens in comparison, and are used for RAG or image processing only.
Model Performance Over Time
Track how model accuracy has evolved over time since their release dates. Each point represents the best performance achieved by that model on GAIA.
Token Pricing Configuration
Adjust token prices to see how they affect the total cost calculations in the leaderboard and plots.
Additional Resources
Getting Started
Want to evaluate your agent on GAIA? Follow our comprehensive guide to get started:
View Documentation