OdysseySpecialist modelResearch preview
Agora-2
PopularityRank 345 of 645
33out of 100-6 vs. last month
based on two data groups, mostly search
Summary
Agora-2 is Odyssey’s interactive multi-agent world model. The research preview creates a shared environment in real time. Up to 20 humans and agents can act inside it, and their actions change the same simulated state.
The public access path is a playable web research preview. Odyssey describes multiple environments, more complex interactions, and longer behavior horizons as part of Agora-2.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 10 published benchmark results
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
Movement branch accuracy (Show measurement, test conditions, and source)
- Source value
- 85.17
- Score
- 85.17
- Metric
- accuracy
- Unit
- %
- Category
- specialist
- Direction
- Higher is better
- Sample
- 2048
- Source type
- vendor-reported
- Evaluator
- Odyssey Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Deployed simulation checkpoint; single-step prediction on 16 batches ×128 held-out recorded examples; averaged across batches.
Blocked-movement branch accuracy (Show measurement, test conditions, and source)
- Source value
- 68.17
- Score
- 68.17
- Metric
- accuracy
- Unit
- %
- Category
- specialist
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Odyssey Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Deployed simulation checkpoint; single-step prediction on obstructed examples; subset count not stated.
Animation mode accuracy (Show measurement, test conditions, and source)
- Source value
- 95.84
- Score
- 95.84
- Metric
- accuracy
- Unit
- %
- Category
- specialist
- Direction
- Higher is better
- Sample
- 2048
- Source type
- vendor-reported
- Evaluator
- Odyssey Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Vendor-reported result. Deployed simulation checkpoint; single-step prediction on 16 batches ×128 held-out recorded examples; averaged across batches.
Single-step predicted animation mode-switch rate (Show measurement, test conditions, and source)
- Source value
- 7.49
- Score
- 7.49
- Metric
- Predicted animation mode-switch rate (recorded reference rate 3.54%)
- Unit
- %
- Benchmark version
- Odyssey Systems Agora-2 report, Table 6; deployed simulation checkpoint
- Category
- single-step simulation diagnostic
- Direction
- No direction published
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2: An Action-Conditioned World Model for Real-Time Multiplayer Environments, Table 6
- Status
- active
- Retrieved at
- 2026-10-18
- Methodology
- Vendor-reported result. Single-step animation-mode prediction from the deployed Agora-2 simulation checkpoint, evaluated on 16 batches of 128 held-out recorded examples, with metrics averaged across batches. Table 6 reports a predicted mode-switch rate of 7.49% and a recorded reference rate of 3.54%. This is a component behavior diagnostic, not a quality ranking or autoregressive rollout result; the immutable checkpoint digest is not published.
- Context
- This is a mode-frequency diagnostic with no monotonic quality direction. It compares the model-predicted transition rate with the recorded reference rate. It is neither accuracy, end-to-end performance, nor autoregressive rollout quality. It is a distinct cell from the same Table 6 evaluation as the three accuracy rows already in the frozen input, not an independent run. No 95% confidence interval is published. The per-cell denominator is not specified; the 16 batches of 128 held-out examples describe the batch setup, not a reported sample count for this metric.
Structured-prefix renderer reconstruction RGB MSE (Show measurement, test conditions, and source)
- Source value
- 0.001279
- Score
- 0.001279
- Metric
- RGB mean squared error (lower is better)
- Unit
- ratio
- Benchmark version
- Odyssey Systems Agora-2 report, Table 3; deployed checkpoint 15,000
- Category
- video-reconstruction-diagnostic
- Direction
- Lower is better
- Sample
- 90
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Renderer component of the deployed Agora-2 system, checkpoint 15,000. Thirty episode-distinct monitoring clips, three sampling seeds each (90 paired evaluations); decoded RGB tensors, 20-frame references at 640×480 and 30 fps. Provider says the collection is not an untouched benchmark. The comparison is between complete systems with different training data and conditioning, not an architecture-isolating test.
- Context
- Vendor-reported renderer-component diagnostic on the provider's monitoring collection, not a matched architecture comparison or an end-to-end agent/world-model score. Lower MSE is better; no confidence interval is published.
Structured-prefix renderer reconstruction PSNR (Show measurement, test conditions, and source)
- Source value
- 30.11
- Score
- 30.11
- Metric
- peak signal-to-noise ratio (PSNR), dB
- Unit
- dB
- Benchmark version
- Odyssey Systems Agora-2 report, Table 3; deployed checkpoint 15,000
- Category
- video-reconstruction-diagnostic
- Direction
- Higher is better
- Sample
- 90
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Renderer component of the deployed Agora-2 system, checkpoint 15,000. Thirty episode-distinct monitoring clips, three sampling seeds each (90 paired evaluations); decoded RGB tensors, 20-frame references at 640×480 and 30 fps. PSNR uses peak value one and is averaged over frames and equally over clips. The collection is provider-used monitoring data, not an untouched benchmark; comparator systems differ in training data and conditioning.
- Context
- Vendor-reported renderer-component diagnostic on the provider's monitoring collection, not a matched architecture comparison or an end-to-end agent/world-model score. PSNR is reported in dB; no confidence interval is published.
Structured-prefix renderer reconstruction RGB MSE at five denoising steps (Show measurement, test conditions, and source)
- Source value
- 0.001294
- Score
- 0.001294
- Metric
- RGB mean squared error (lower is better)
- Unit
- ratio
- Benchmark version
- Odyssey Systems Agora-2 report, Table 9; deployed checkpoint 15,000; 5 denoising steps
- Category
- video-reconstruction-diagnostic
- Direction
- Lower is better
- Sample
- 30
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Within-model renderer diagnostic at five denoising steps on 30 episode-distinct monitoring clips, one sampling seed per clip. The report holds weights, conditioning, visual history, precision, and cache-update noise fixed. This collection is provider-used monitoring data, not an untouched benchmark.
- Context
- Vendor-reported renderer-component sensitivity result from one seed; not a cross-model comparison or end-to-end agent/world-model score. Lower MSE is better; no confidence interval is published.
Structured-prefix renderer reconstruction PSNR at five denoising steps (Show measurement, test conditions, and source)
- Source value
- 30.12
- Score
- 30.12
- Metric
- peak signal-to-noise ratio (PSNR), dB
- Unit
- dB
- Benchmark version
- Odyssey Systems Agora-2 report, Table 9; deployed checkpoint 15,000; 5 denoising steps
- Category
- video-reconstruction-diagnostic
- Direction
- Higher is better
- Sample
- 30
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Within-model renderer diagnostic at five denoising steps on 30 episode-distinct monitoring clips, one sampling seed per clip. PSNR is averaged over frames and clips. The report holds weights, conditioning, visual history, precision, and cache-update noise fixed; the collection was used by the provider for monitoring.
- Context
- Vendor-reported renderer-component sensitivity result from one seed; not a cross-model comparison or end-to-end agent/world-model score. PSNR is reported in dB; no confidence interval is published.
Structured-prefix renderer reconstruction RGB MSE at ten denoising steps (Show measurement, test conditions, and source)
- Source value
- 0.001319
- Score
- 0.001319
- Metric
- RGB mean squared error (lower is better)
- Unit
- ratio
- Benchmark version
- Odyssey Systems Agora-2 report, Table 9; deployed checkpoint 15,000; 10 denoising steps, one-seed subset
- Category
- video-reconstruction-diagnostic
- Direction
- Lower is better
- Sample
- 30
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Within-model renderer diagnostic at ten denoising steps on 30 episode-distinct monitoring clips, one sampling seed per clip. The report says these ten-step outputs are reused from the complete-system comparison, so this is a subset and not an independent rerun. The collection is provider-used monitoring data.
- Context
- Vendor-reported renderer-component sensitivity result. The 30 ten-step outputs are reused from the larger complete-system comparison; do not count this subset as an independent run. Lower MSE is better. No confidence interval is published.
Structured-prefix renderer reconstruction PSNR at ten denoising steps (Show measurement, test conditions, and source)
- Source value
- 30.08
- Score
- 30.08
- Metric
- peak signal-to-noise ratio (PSNR), dB
- Unit
- dB
- Benchmark version
- Odyssey Systems Agora-2 report, Table 9; deployed checkpoint 15,000; 10 denoising steps, one-seed subset
- Category
- video-reconstruction-diagnostic
- Direction
- Higher is better
- Sample
- 30
- Source type
- vendor-reported
- Evaluator
- Odyssey Systems, Agora-2 technical report
- Status
- active
- Retrieved at
- 2026-10-16
- Methodology
- Vendor-reported result. Within-model renderer diagnostic at ten denoising steps on 30 episode-distinct monitoring clips, one sampling seed per clip. PSNR is averaged over frames and clips. The report says these ten-step outputs are reused from the complete-system comparison, so this is a subset and not an independent rerun.
- Context
- Vendor-reported renderer-component sensitivity result. The 30 ten-step outputs are reused from the larger complete-system comparison; do not count this subset as an independent run. PSNR is in dB. No confidence interval is published.
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
15 of 15 models
As of October 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 9, 2026
Agora-2TeleOCR 39Kumo-Tabular 38LensVLM-9B 32AstaBrief-8B 30
Method changes in the timeline, a jump at these points is no market event:
- 1November 1, 2026: Value only with search or usage data, 30-day attention
- 2November 1, 2026: Method version 4 (common scale, shrinkage)
- 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)
What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.
Similar models
More models of the same kind as Agora-2, chosen by model line, developer, size, price and release date.
- SAM 3Meta · popular in this category
- EmbeddingGemma 2Google DeepMind · popular in this category
- Unlimited OCR 2Baidu · popular in this category
- Qwen-Drive-1.0-4BQwen · popular in this category
- TeleOCRTeleAI · popular in this category
- Kumo-TabularNVIDIA · popular in this category
- LensVLM-9BApple · popular in this category
All specifications
Architecture
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- Shutdown
- None announced
- no source
Further details
Sources and data date
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 15, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not published | long-horizon-interactive-quality Note Report says long-horizon visual quality, cross-view agreement, action adherence and agent-policy performance remain unevaluated. |
| Not published | long-horizon-interactive-quality Note The official report still says long-horizon visual quality, cross-view agreement, action adherence, and agent-policy performance are unevaluated. Its figure is qualitative, not a benchmark score. |
| Additional source | Odyssey announcement (retrieved September 26, 2026) · Odyssey announcement · Editorial description reviewed October 2, 2026 |
| Additional source | Odyssey Agora-2 technical report (retrieved October 15, 2026; October 16, 2026; October 18, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Ollama: cumulative downloads for the popularity index |
| Additional source | Replicate: cumulative runs for the popularity index |
| Additional source | Google Trending Now: limited feed selection for Germany and the US |
| Additional source | Creative Commons Attribution 4.0 |