General AI Assistants (GAIA)

GAIA evaluates AI models on real-world tasks that are conceptually simple for humans but require multi-step reasoning, web browsing, tool use, and multimodal understanding for AI. Tasks span three difficulty levels and test practical assistant capabilities rather than academic knowledge.

How BenchLM shows GAIA right now

BenchLM is tracking GAIA in the local dataset, but exact-source verification records for these rows are still being attached. To avoid a blank benchmark page, BenchLM shows the current tracked rows below as a display-only reference table.

These tracked rows are useful for inspection and spot-checking, but until exact-source attachments are completed they should not be treated as fully verified public benchmark rows.

26 tracked modelsLocal tracked rowsAwaiting exact-source attachmentsDisplay only

Tracked score on GAIA — April 7, 2026

BenchLM mirrors the published tracked score view for GAIA. Claude Mythos Preview leads the public snapshot at 52.3% , followed by GPT-5.4 Pro (50.5%) and GPT-5.4 (48.2%). BenchLM does not use these results to rank models overall.

26 modelsAgentic12% of category scoreRefreshingUpdated April 7, 2026

The published GAIA snapshot is tightly clustered at the top: Claude Mythos Preview sits at 52.3%, while the third row is only 4.1 points behind. The broader top-10 spread is 12.6 points, so the benchmark still separates strong models even when the leaders cluster.

26 models have been evaluated on GAIA. The benchmark falls in the Agentic category. This category carries a 22% weight in BenchLM.ai's overall scoring system. Within that category, GAIA contributes 12% of the category score, so strong performance here directly affects a model's overall ranking.

About GAIA

Year

2024

Tasks

466

BenchLM freshness & provenance

Version

GAIA 2024

Refresh cadence

Annual

Staleness state

Refreshing

Question availability

Public benchmark set

Refreshing

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Tracked score table (26 models)

#1
Claude Mythos Previewclaude-mythos-preview
52.3%
#2
GPT-5.4 Progpt-5-4-pro
50.5%
#3
GPT-5.4gpt-5-4
48.2%
#4
Claude Opus 4.6claude-opus-4-6
47.8%
#5
Gemini 3.1 Progemini-3-1-pro
46.1%
#6
Claude Sonnet 4.6claude-sonnet-4-6
45.5%
#7
GPT-5 minigpt-5-mini
44.8%
#8
GPT-5 (high)gpt-5-high
42.1%
#9
GPT-5.2gpt-5-2
40.3%
#10
Grok 4.1grok-4-1
39.7%
#11
Gemini 3 Progemini-3-pro
38.5%
#12
Kimi K2.5 (Reasoning)kimi-k2-5-reasoning
38.1%
#13
Qwen3.6 Plusqwen3-6-plus
37.4%
#14
o4-mini (high)o4-mini-high
36.8%
#15
Kimi K2.5kimi-k2-5
36.5%
#16
Qwen3.5 397B (Reasoning)qwen3-5-397b-reasoning
35.6%
#17
Gemini 3 Flashgemini-3-flash
35.2%
#18
DeepSeek V3.2 (Thinking)deepseek-v3-2-thinking
34.8%
#19
Llama 4 Behemothllama-4-behemoth
34.2%
#20
GLM-5 (Reasoning)glm-5-reasoning
33.8%
#21
Qwen3.5 397Bqwen3-5-397b
33.2%
#22
DeepSeek V3.2deepseek-v3-2
32.4%
#23
GLM-5glm-5
31.5%
#24
Mistral Large 3mistral-large-3
30.8%
#25
Step 3.5 Flashstep-3-5-flash
29.4%
#26
Llama 4 Maverickllama-4-maverick
28.6%

FAQ

What does GAIA measure?

GAIA evaluates AI models on real-world tasks that are conceptually simple for humans but require multi-step reasoning, web browsing, tool use, and multimodal understanding for AI. Tasks span three difficulty levels and test practical assistant capabilities rather than academic knowledge.

Which model leads the published GAIA snapshot?

Claude Mythos Preview currently leads the published GAIA snapshot with a tracked score of 52.3%. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on GAIA?

26 AI models are included in BenchLM's mirrored GAIA snapshot, based on the public leaderboard captured on April 7, 2026.

Last updated: April 7, 2026 · mirrored from the public benchmark leaderboard

Weekly LLM Benchmark Digest

Get notified when new models drop, benchmark scores change, or the leaderboard shifts. One email per week.

Free. No spam. Unsubscribe anytime. We only store derived location metadata for consent routing.