Nous: Hermes 4 70B

Text input Text output
Author's Description

Hermes 4 70B is a hybrid reasoning model from Nous Research, built on Meta-Llama-3.1-70B. It introduces the same hybrid mode as the larger 405B release, allowing the model to either respond directly or generate explicit <think>...</think> reasoning traces before answering. Users can control the reasoning behaviour with the `reasoning` `enabled` boolean. [Learn more in our docs](https://openrouter.ai/docs/use-cases/reasoning-tokens#enable-reasoning-with-default-config) This 70B variant is trained with the expanded post-training corpus (~60B tokens) emphasizing verified reasoning data, leading to improvements in mathematics, coding, STEM, logic, and structured outputs while maintaining general assistant performance. It supports JSON mode, schema adherence, function calling, and tool use, and is designed for greater steerability with reduced refusal rates.

Key Specifications
Cost
$$
Context
131K
Parameters
70B
Released
Aug 26, 2025
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Top Logprobs Stop Top P Seed Frequency Penalty Max Tokens Reasoning Logprobs Presence Penalty Include Reasoning Logit Bias Temperature
Features

This model supports the following features:

Reasoning
Performance Summary

Nous: Hermes 4 70B, built on Meta-Llama-3.1-70B, demonstrates strong overall performance, particularly excelling in reliability with a 94% success rate, indicating consistent and usable responses. The model performs among the fastest models, ranking in the 69th percentile for speed, and offers competitive pricing, placing in the 74th percentile for cost-effectiveness. In terms of specific benchmarks, Hermes 4 70B shows notable strengths in Ethics (99.0% accuracy) and Email Classification (98.0% accuracy), suggesting robust performance in tasks requiring ethical judgment and precise categorization. General Knowledge also shows solid performance at 96.5% accuracy. However, the model exhibits weaknesses in Coding (26th percentile accuracy) and Reasoning (29th percentile accuracy), indicating areas for potential improvement in complex problem-solving and programming tasks. Instruction Following also presents a moderate challenge at 55.1% accuracy. Its hybrid reasoning mode and support for JSON, schema adherence, and function calling are key features enhancing its steerability and utility for structured outputs.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.13
Completion $0.4

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
Nebius
Nebius | nousresearch/hermes-4-70b 131K $0.13 / 1M tokens $0.4 / 1M tokens
Chutes
Chutes | nousresearch/hermes-4-70b 131K $0.0933 / 1M tokens $0.373 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Free Executions Accuracy Cost Duration
Other Models by nousresearch