Hermes 4 70B
Unavailable in this round
Language modelActive
Hermes 4 70B
Nous Research
- Released
- August 18, 2025
- Data date
- September 8, 2026
- Predecessor
- Hermes 3 70B (Official documentation)
- Price
- Historical rates under Costs
- Input
- Text
- Output
- Text
Hermes 4 70B combines an explicit reasoning mode with direct answers and structured JSON on top of Llama 3.1 70B.
Nous Research describes roughly 5 million samples and 60 billion tokens in post-training compared with Hermes 3, along with stronger steerability and schema adherence. The technical report gives 56 billion tokens for a different training setup, so those figures are not interchangeable.
Specifications and access
| Specification | Value and source |
|---|---|
| Weights | Open weightsSource |
| Access | LocalSource |
| Parameters | 70BSource |
| Context window | 131,072 tokensSource |
| License | Llama 3.1 Community LicenseSource |
| Parameters | 70 BSource |
| Architecture | dense Llama 3.1 decoder-only transformer, 80 layersSource |
| Input | TextSource |
| Output | TextSource |
| Reasoning mode | hybrid mode; model can emit explicit <think> reasoning or answer in non-reasoning modeSource |
| Tool calling | YesSource |
| Structured output | YesSource |
| Post-training samples | 5,000,000Source |
| Post-training tokens | 60 B (The technical report separately states 56B training tokens for its reported setup.)Source |
| Training tokens | 56 B (Do not silently merge this report-wide figure with the card's approximately 60B post-training description.)Source |
| Training context | 16,384 tokensSource |
Page 1 of 4
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.
MATH500 (reasoning) (Show measurement, test conditions, and source)
- Source value
- 95.6
- Score
- 95.6
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
MATH500 (non-reasoning) (Show measurement, test conditions, and source)
- Source value
- 71
- Score
- 71
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
AIME24 (reasoning) (Show measurement, test conditions, and source)
- Source value
- 73.5
- Score
- 73.5
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
AIME24 (non-reasoning) (Show measurement, test conditions, and source)
- Source value
- 9.5
- Score
- 9.5
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
AIME25 (reasoning) (Show measurement, test conditions, and source)
- Source value
- 67.4
- Score
- 67.4
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
AIME25 (non-reasoning) (Show measurement, test conditions, and source)
- Source value
- 7.3
- Score
- 7.3
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
GPQA (reasoning) (Show measurement, test conditions, and source)
- Source value
- 66.1
- Score
- 66.1
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
GPQA (non-reasoning) (Show measurement, test conditions, and source)
- Source value
- 33.3
- Score
- 33.3
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
BBH (reasoning) (Show measurement, test conditions, and source)
- Source value
- 87.8
- Score
- 87.8
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
BBH (non-reasoning) (Show measurement, test conditions, and source)
- Source value
- 80.5
- Score
- 80.5
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
LCBv6 (reasoning) (Show measurement, test conditions, and source)
- Source value
- 50.5
- Score
- 50.5
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
LCBv6 (non-reasoning) (Show measurement, test conditions, and source)
- Source value
- 25.5
- Score
- 25.5
- Metric
- percent
- Unit
- %
- Category
- mixed
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Hermes 4 Technical Report Table 4
- Status
- active
- Retrieved at
- 2026-09-08
- Methodology
- Vendor-reported result. Official report Table 4; each row preserves the R (reasoning) or N (non-reasoning) column.
Page 1 of 4
Our tests
SVG short test
One SVG prompt and fixed output settings. Captures are shown only with 0 reported reasoning tokens or documented lack of reasoning controls. Identical test conditions do not make the models' internal compute identical.
Popularity
38.5 / 100
Predecessor and successor
Sources and data date
| Type | Evidence and data date |
|---|---|
| Input | Documented modalities. Retrieved September 8, 2026. |
| Output | Documented modalities. Retrieved September 8, 2026. |
| Research status | Research date September 8, 2026. 4 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | active-parameters-billions Original wording The exact model is a dense 70B model; no separate active-parameter count applies. |
| Not found in the inspected sources | knowledge-cutoff Original wording The exact card/report does not publish a knowledge cutoff. |
| Additional source | API pricing |
| Additional source | Hermes 4 Technical Report Table 4 (retrieved September 8, 2026) |
| Additional source | Artificial Analysis (retrieved September 8, 2026) |
| Additional source | Model metadata (retrieved September 8, 2026) |
| Additional source | Nous Research Hermes 4 technical report · Editorial description reviewed October 2, 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 | Creative Commons Attribution 4.0 |