Text and chat · Meta (Hermes 3: Nous Research)

Llama and Hermes 3

Meta's older open Llama models and the Hermes 3 fine-tune, in private mode. Newer families do better for the price.

Private Moderate cost
Latest versionLlama 3.3 70BDecember 6, 2024
Artificial Analysis Intelligence Index8 points#86 of 90 on Artificial Analysis Intelligence Index
Price$1.259 per million tokens written$0.000881 a page written
Key spec128K tokens of context

What sets it apart

Meta’s Llama line made open weights mainstream, and Hermes 3 shows what a community fine-tune adds on top. In the catalog these are older, dense, text-only models: predictable and cheap to run, but newer families score higher.

  • Dense open weights from 3B to 405B
  • Llama 3.3 70B matches 3.1 70B at lower cost
  • Hermes 3 adds steerable system prompts

Strengths and limits

Where it shines

  • Private mode, well-known behavior
  • Hermes 3 follows your system prompt and persona closely

Think twice when

  • Older generation: no vision, no reasoning mode
  • Hermes 3 has no tool calling, so none of Sub Rosa's agent tools
  • Llama 3.3 70B costs more than newer, stronger models
  • 128K token context
  • Replies capped at 4K tokens (Llama)

Use cases

Cheap bulk text chores

Rewrite each of these 20 product descriptions in a friendly tone, under 40 words each. Keep the product names unchanged.

Llama 3.2 3B costs very little and handles simple rewriting.

Steerable character writing

You are Captain Mara, a sarcastic ship’s pilot. Stay in character and describe our arrival at a storm-wrecked harbor in four sentences.

Hermes 3 follows a persona set in the system prompt closely.

General assistant on a budget

Summarize this email thread in five bullet points and list the decisions that are still open.

Llama 3.3 70B is a solid mid-size all-rounder with tool calling.

Against its rivals

  • GemmaNewer Google open models that read images, where Llama 3.3 is text only.#71 of 90 on Artificial Analysis Intelligence Index (17 points)
  • Mistral SmallMistral Small is far cheaper, reads images and has a 256K context.#87 of 90 on Artificial Analysis Intelligence Index (8 points)
  • Qwen open modelsQwen open models offer reasoning and larger context at similar prices.#23 of 90 on Artificial Analysis Intelligence Index (40 points)
Compare side by side ↗

Version history

Each version, dated, with what it changed over the one before.

  1. Llama 3.1Earlier, not in Sub Rosa
    • Introduced the 405B flagship and 128K context
    • Eight supported languages and better tool use
    Source : ai.meta.com
  2. Hermes 3 (Llama 3.1 405B)
    • Nous Research fine-tune of Llama 3.1 405B with function calling and strong steerability through system prompts
    Source : huggingface.co
  3. Llama 3.2 3B
    • Small models distilled from 3.1 8B and 70B, built for phones and edge devices
    • Quantized variants for faster on-device decoding
    Source : huggingface.co
  4. Llama 3.3 70BLatest
    • 70B model improved over 3.1 70B: IFEval 92.1 vs 87.5, MATH 77.0 vs 68.0
    • Refined multilingual training and fewer false refusals
    Source : huggingface.co

Benchmarks

How to read these scores: what a benchmark tells you.

Artificial Analysis Intelligence Index: #86 of 90 models measured in the catalog. One score averaging about ten independent tests of reasoning, knowledge, math, coding and agent work, all run the same way for every model. Higher is better. Each model is listed at its highest reasoning effort. Two or three points apart is a close call; ten points is a different class of model.

Artificial Analysis Intelligence IndexHigher is better · Artificial Analysis, October 4, 2026
  1. 1. Claude Opus 5.5Claude58
  2. 2. Claude Sonnet 5.5Claude56
  3. 3. Claude Fable 5.1Claude53
  4. 4. GPT-6 AstraGPT53
  5. 5. Gemini 4 ArgonGemini53
  6. 6. GPT-6.1 SolGPT52
  7. 7. Muse WonderMuse52
  8. 8. Claude Opus 5Claude51
  9. 9. Claude Fable 5Claude50
  10. 10. GPT-6 SolGPT48
  11. 86. Llama 3.3 70BLlama and Hermes 38
See the numbers
Modelpoints
1. Claude Opus 5.558
2. Claude Sonnet 5.556
3. Claude Fable 5.153
4. GPT-6 Astra53
5. Gemini 4 Argon53
6. GPT-6.1 Sol52
7. Muse Wonder52
8. Claude Opus 551
9. Claude Fable 550
10. GPT-6 Sol48
86. Llama 3.3 70B8

All 3 models in this family

ModelUsePrivacyPrice
Llama 3.3 70Bllama-3.3-70bTextPrivate$1.259 per million tokens written$0.000881 a page written
Llama 3.2 3Bllama-3.2-3bTextPrivate$0.27 per million tokens written$0.000189 a page written
Hermes 3 Llama 3.1 405bhermes-3-llama-3.1-405bTextPrivate$1.349 per million tokens written$0.000944 a page written

Specifications

SpecificationLlama 3.3 70BLlama 3.2 3BHermes 3 405B
ReleasedDecember 6, 2024September 25, 2024August 2024
Parameters70B3.21B406B
WeightsOpen, Llama 3.3 Community LicenseOpen, Llama 3.2 Community LicenseOpen, Llama 3.1 Community License
Acceptstexttexttext
Producestexttexttext
Context128K tokens128K tokens128K tokens
Longest answer4K tokens4K tokens16K tokens
Languages8 languages8 languages·
AbilitiesUses Sub Rosa’s toolsUses Sub Rosa’s tools·
Price$1.259 per million tokens written$0.27 per million tokens written$1.349 per million tokens written
PrivacyPrivatePrivatePrivate

Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.