Language modelLimited access
Gemini 4 Argon
Google DeepMind
- Released
- -
- Data date
- October 3, 2026
Gemini 4 Argon expands Google’s maximum output limit from 64K to 1 million tokens, allowing much longer reasoning on extended coding and research tasks. Google announced the model on September 30, 2026 and also emphasizes finding, validating, and patching security vulnerabilities.
Access initially goes to trusted cyber defenders through the Fairwind Program. Google plans a broader release for developers, enterprises, and consumers but has not specified a date.
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.
DeepSWE v1.1 (Show measurement, test conditions, and source)
- Source value
- 77.9
- Score
- 77.9
- Metric
- DeepSWE v1.1
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- blog.google
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Announcement; detailed sampling, tool and reasoning configuration not specified in the inspected text
- Context
- Source-specific observation; it is not a shared comparison cohort. These are numeric source-specific claims, not verified independent leaderboard records. Qualitative Vals, Harvey, Gray Swan and internal-security claims remain without numeric scores. Complete charts and independent run configuration were not extracted.
AutomationBench (Show measurement, test conditions, and source)
- Source value
- 51.3
- Score
- 51.3
- Metric
- AutomationBench
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- blog.google
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Announcement; exact dataset revision and run configuration not specified
- Context
- Source-specific observation; it is not a shared comparison cohort. These are numeric source-specific claims, not verified independent leaderboard records. Qualitative Vals, Harvey, Gray Swan and internal-security claims remain without numeric scores. Complete charts and independent run configuration were not extracted.
LVBench (Show measurement, test conditions, and source)
- Source value
- 91.7
- Score
- 91.7
- Metric
- LVBench
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- blog.google
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Announcement; long-video understanding; detailed run configuration not specified
- Context
- Source-specific observation; it is not a shared comparison cohort. These are numeric source-specific claims, not verified independent leaderboard records. Qualitative Vals, Harvey, Gray Swan and internal-security claims remain without numeric scores. Complete charts and independent run configuration were not extracted.
CWE-bench v1 (Show measurement, test conditions, and source)
- Source value
- 68
- Score
- 68
- Metric
- CWE-bench v1
- Unit
- %
- Category
- source-specific
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- blog.google
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Announcement; vulnerability remediation; separate from CWE-bench v0 and internal security evaluations
- Context
- Source-specific observation; it is not a shared comparison cohort. These are numeric source-specific claims, not verified independent leaderboard records. Qualitative Vals, Harvey, Gray Swan and internal-security claims remain without numeric scores. Complete charts and independent run configuration were not extracted.
Sources and data date
Every statement links to its underlying documentation or leaderboard.
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 4, 2026. 1 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Additional source | Google Gemini 4 Argon announcement (retrieved October 3, 2026; October 4, 2026) · Google Gemini 4 Argon announcement · Editorial description reviewed October 4, 2026 |