Agnes 3.0 Flash
Agnes AI
- Released
- -
- Data date
- October 7, 2026
- Input
- Text and image URLs
- Output
- Text
Agnes 3.0 Flash is Agnes AI's API model for coding agents and multi-step tool workflows. Its documentation lists text and image URLs as inputs, text output, and up to 65,536 output tokens.
Agnes AI explicitly separates the production API checkpoint from the open preview weights. Its pages give different context limits. The API documentation says 512K, while the model card says 1M for the newer API checkpoint.
Specifications and access
| Specification | Value and source |
|---|---|
| Provider model ID | agnes-3.0-flashSource |
| Access | Agnes APISource |
| Input modalities | Text and image URLsSource |
| Output modalities | TextSource |
| Context in API documentation | 512K tokensSource |
| Context in model card | 1M tokens for the newer API checkpoint; differs from the API documentationSource |
| Checkpoint distinction | Production API checkpoint; not the open 33B preview checkpoint.Source |
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.
AA output speed, provider-reported September 11, 2026 (Show measurement, test conditions, and source)
- Source value
- 252.7
- Score
- 252.7
- Metric
- provider-reproduced output tokens per second
- Unit
- tokens/s
- Benchmark version
- AA provider speed snapshot, 2026-09-11
- Category
- performance
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Agnes AI model documentation
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Agnes AI documentation reproduces an Artificial Analysis output-speed value for the exact production/API model agnes-3.0-flash as of September 11, 2026. The provider does not disclose the input workload, sample count or timing aggregation. This is a dated vendor reproduction, not a currently verified independent API run.
- Context
- Historical provider-reproduced throughput, without the original independent timing protocol. It must not be interpreted as the current API speed or as a quality benchmark.
Published prices
- Provider-listed API price
- List price per 1 million tokens: $0.005 cache hit, $0.05 input, $0.15 output. The current promotion lists all three at $0; no end date is stated.Source
Sources and data date
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved October 7, 2026. |
| Output | Documented modalities. Retrieved October 7, 2026. |
| Research status | Research date October 7, 2026. 6 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Unresolved | artificial-analysis-intelligence-index Original wording Artificial Analysis marks the Agnes 3.0 Flash Intelligence Index v4.3 value 36 as an estimate while independent evaluation is ongoing. Agnes AI documentation repeats that estimate as of 2026-09-11. It is not included as a completed benchmark measurement. |
| Unresolved | lumina-conditional-indices Original wording Lumina publishes conditional category indices for the exact API identity, but identifies Artificial Analysis as the source and says missing evidence is predicted. It does not disclose the exact measured source inputs behind these indices, so no Lumina index is included as a benchmark result. |
| Unresolved | completed-independent-task-results Original wording The AA comparison page was read earlier, but current direct model and comparison pages return HTTP 404. Agnes AI explicitly says independent evaluation is still ongoing. The component values cannot be confirmed as completed independent runs and are omitted; only the separately documented historical vendor-reproduced throughput remains. |
| Additional source | Agnes 3.0 Flash Preview model card (retrieved October 7, 2026) · Agnes 3.0 Flash Preview model card · Editorial description reviewed October 7, 2026 |
| Additional source | Model metadata (retrieved October 7, 2026) |
| Additional source | Model metadata (retrieved October 7, 2026) |