Agents · Benchmark profile
A benchmark for general assistants that must reason, browse, use tools and combine modalities.
Data verified 15 July 2026 · Methodology 1.2.0
Benchmark score on GAIA
Claude Fable 5 leads at 52.3%, followed by Claude Mythos 5 (52.3%) and GPT-5.4 (48.2%).
Visual analysis
Three views of the same sourced leaderboard: model placement, the score curve, and descriptive provider averages. These visuals never alter ranking calculations.
13 ranked models · higher is better
| Model | Rank | Provider | Score |
|---|---|---|---|
| Claude Fable 5 | #1 | Anthropic | 52.3% |
| Claude Mythos 5 | #2 | Anthropic | 52.3% |
| GPT-5.4 | #3 | OpenAI | 48.2% |
| Claude Opus 4.6 | #4 | Anthropic | 47.8% |
| Gemini 3.1 Pro Preview | #5 | 46.1% | |
| Claude Sonnet 4.6 | #6 | Anthropic | 45.5% |
| Grok 4.1 | #7 | xAI | 39.7% |
| Gemini 3 Pro | #8 | 38.5% | |
| Qwen3.6 Plus | #9 | Alibaba Cloud | 37.4% |
| Kimi K2.5 | #10 | Moonshot AI | 36.5% |
| Gemini 3 Flash | #11 | 35.2% | |
| GLM-5 | #12 | Z.AI | 31.5% |
| Mistral Large 3 | #13 | Mistral AI | 30.8% |
Shows how quickly scores separate down the field
| Model | Rank | Provider | Score |
|---|---|---|---|
| Claude Fable 5 | #1 | Anthropic | 52.3% |
| Claude Mythos 5 | #2 | Anthropic | 52.3% |
| GPT-5.4 | #3 | OpenAI | 48.2% |
| Claude Opus 4.6 | #4 | Anthropic | 47.8% |
| Gemini 3.1 Pro Preview | #5 | 46.1% | |
| Claude Sonnet 4.6 | #6 | Anthropic | 45.5% |
| Grok 4.1 | #7 | xAI | 39.7% |
| Gemini 3 Pro | #8 | 38.5% | |
| Qwen3.6 Plus | #9 | Alibaba Cloud | 37.4% |
| Kimi K2.5 | #10 | Moonshot AI | 36.5% |
| Gemini 3 Flash | #11 | 35.2% | |
| GLM-5 | #12 | Z.AI | 31.5% |
| Mistral Large 3 | #13 | Mistral AI | 30.8% |
Average within the visible cohort · descriptive only
| Provider | Models | Average score |
|---|---|---|
| Anthropic | 4 | 49.5% |
| OpenAI | 1 | 48.2% |
| 3 | 39.9% | |
| xAI | 1 | 39.7% |
| Alibaba Cloud | 1 | 37.4% |
| Moonshot AI | 1 | 36.5% |
| Z.AI | 1 | 31.5% |
| Mistral AI | 1 | 30.8% |
One best score per model · higher is better
| Rank | Model | Provider | License | Score |
|---|---|---|---|---|
| #1 | Claude Fable 5 claude-fable-5 | Anthropic | closed | 52.3% |
| #2 | Claude Mythos 5 claude-mythos-5 | Anthropic | closed | 52.3% |
| #3 | GPT-5.4 gpt-5.4 | OpenAI | closed | 48.2% |
| #4 | Claude Opus 4.6 claude-opus-4-6 | Anthropic | closed | 47.8% |
| #5 | Gemini 3.1 Pro Preview gemini-3.1-pro-preview | closed | 46.1% | |
| #6 | Claude Sonnet 4.6 claude-sonnet-4-6 | Anthropic | closed | 45.5% |
| #7 | Grok 4.1 grok-4-1 | xAI | closed | 39.7% |
| #8 | Gemini 3 Pro gemini-3-pro | closed | 38.5% | |
| #9 | Qwen3.6 Plus qwen3-6-plus | Alibaba Cloud | closed | 37.4% |
| #10 | Kimi K2.5 kimi-k2.5 | Moonshot AI | — | 36.5% |
| #11 | Gemini 3 Flash gemini-3-flash | closed | 35.2% | |
| #12 | GLM-5 glm-5 | Z.AI | open | 31.5% |
| #13 | Mistral Large 3 mistral-large-2512 | Mistral AI | open | 30.8% |
The top of this snapshot is led by Claude Fable 5 at 52.3%; third place is 4.1 points behind. The top-10 spread is 15.8 points.
About GAIA
A benchmark for general assistants that must reason, browse, use tools and combine modalities. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.
Open benchmark source ↗FAQ
A benchmark for general assistants that must reason, browse, use tools and combine modalities.
Claude Fable 5 by Anthropic currently leads with 52.3%.
13 models in the LuminaBench cohort have a qualifying score on this benchmark.
Yes. This family is ranking-weighted at 1.6% of its capability category. Claude Fable 5 is overall #2.
Related
