Benchmark Almanac

Meta Agents Research Environments (ARE) / Gaia2

GPT-5 42.1% leads since Sep 2025 Fading

Simulated app environments for agents with time and events. Hosts the Gaia2 benchmark.

Gaia2 has 800 dynamic scenarios across 10 universes. In 2026 Meta added Gaia2 CLI, a container stack that runs agents such as OpenClaw and mini-SWE-agent, and the Omnilingual-Gaia2 translation.

Made by Meta FAIR Released Sep 2025 General assistants All tasks public Last in a launch post Aug 2026 , dropped before it saturated Last model added Oct 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
facebookresearch/meta-agents-research-environments 575 stars Paper Dataset Leaderboard
13
Models scored on its own leaderboard
2
Launch posts list it, from 2 labs
2
System cards report it, from 2 labs

GPT-5 leads with 42.1% among models released by Sep 2025

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in Sep 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

13 models. Scroll sideways for the rest.

  1. 42.1 GPT-5
  2. 37.8 Claude Sonnet 4
  3. 25.8 Gemini 2.5 Pro
  4. 23.1 DeepSeek-V3.1-Terminus
  5. 21.9 DeepSeek-V3.1
  6. 20.1 Kimi K2
  7. 15.7 Grok 4
  8. 15.7 Qwen3-235B-A22B-Thinking
  9. 13.7 gpt-oss-120b
  10. 11.6 Qwen3-235B-A22B
  11. 7.40 GPT-4o
  12. 7.40 Llama 4 Maverick
  13. 4.40 Llama 3.3 70B Instruct
  1. 1 GPT-5 high 42.1%
  2. 2 Claude Sonnet 4 thinking 37.8%
  3. 3 Gemini 2.5 Pro Jun 2025 25.8%
  4. 4 DeepSeek-V3.1-Terminus 23.1%
  5. 5 DeepSeek-V3.1 21.9%
  6. 6 Kimi K2 Sep 2025 20.1%
  7. 7 Grok 4 15.7%
  8. 8 Qwen3-235B-A22B-Thinking Jul 2025, thinking 15.7%
  9. 9 gpt-oss-120b high 13.7%
  10. 10 Qwen3-235B-A22B 11.6%
  11. 11 GPT-4o Mar 2025 7.40%
  12. 12 Llama 4 Maverick 7.40%
  13. 13 Llama 3.3 70B Instruct 4.40%
#ModelScoreSettingDateSource
1 GPT-5 42.1% high Aug 2025 Official leaderboard
2 Claude Sonnet 4 37.8% thinking May 2025 Official leaderboard
3 Gemini 2.5 Pro Jun 2025 25.8% Jun 2025 Official leaderboard
4 DeepSeek-V3.1-Terminus 23.1% Sep 2025 Official leaderboard
5 DeepSeek-V3.1 21.9% Aug 2025 Official leaderboard
6 Kimi K2 Sep 2025 20.1% Sep 2025 Official leaderboard
7 Grok 4 15.7% Jul 2025 Official leaderboard
8 Qwen3-235B-A22B-Thinking Jul 2025 15.7% thinking Jul 2025 Official leaderboard
9 gpt-oss-120b 13.7% high Aug 2025 Official leaderboard
10 Qwen3-235B-A22B 11.6% Apr 2025 Official leaderboard
11 GPT-4o Mar 2025 7.40% Mar 2025 Official leaderboard
12 Llama 4 Maverick 7.40% Apr 2025 Official leaderboard
13 Llama 3.3 70B Instruct 4.40% Dec 2024 Official leaderboard

Labs report scores for 3 models in their cards. We plot every source in the charts below.

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

OpenAI Ant Group
Own leaderboard Not tested since Sep 2025
GPT-5 42.1%
Lab claims 2 leaders
Ring-2.6-1T 77.9%

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Ring-2.6-1T GPT-5 Claude Sonnet 4 Qwen3-235B-A22B GPT-4o 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Ring-2.6-1T GPT-5

Labs headline it less often now, last in Aug 2026

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Ant Group
Meta
DateLaunch postScoreAs the post writes it
10 Aug 2026 Muse Glimmer 43.3 Gaia2
10 Feb 2026 Ring-2.5-1T 75 Gaia2-search (Mean@3)

2 system cards report it, from 2 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
19 Oct 2026 Muse Glimmer technical report 43.3 GAIA2, high (Table 2) Capabilities
13 Jun 2026 Ring-2.6-1T technical report 77.9 GAIA-2 Search (Pass@1, 3 runs), xhigh (Table 6). High gives 75.40 Capabilities

More in Tool use

All 125 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Muse Spark 1.1 88.1%
  • BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Claw-Eval In use 300 human-verified agent tasks (service orchestration, multimodal, multi-turn dialogue) audited via traces and snapshots. Rubric scores with Pass^3 consistency. Claude Sonnet 4.6 81.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%