Llama 3.3

Available · Language

Llama 3.3 is a text-only multilingual 70B model from Meta, released on 6 December 2024 as open weights under the custom Llama 3.3 Community License, in an instruction-tuned version only. Meta says it improves on Llama 3.1 70B; in April 2025 Meta added a Llama 3.3 8B model for fine-tuning in its Llama API preview. [1] [2] [3] [4]Introducing Llama API in preview Primary source

Timeline of Llama 3.3 →

Claims and evidence

  • Released Other sources give: [5]README, Llama Models table, Launch date Launch date given in the table of Meta's llama-models README; the model cards say December 6, 2024. Primary source[1]Model Release Date [3]model card, Model Release Date; license, Version Release Date
  • Status AvailableApplies to Llama 3.3 70B, whose weights are still listed in Meta's Hugging Face organisation (gated, license acceptance required) as seen on 2026-10-01. Meta publishes no deprecation page for Llama models. See the 8B variant for its own status. Primary source[3] [2]Download the Model
  • Successor of Llama 3.2 EditorialEditorial link: next point release in the Llama 3 line after Llama 3.2. Meta lists Llama 3, 3.1, 3.2 and 3.3 in that order and compares 3.3 with Llama 3.1 70B and, for text-only use, Llama 3.2 90B, without calling it a successor. Primary source[2]Introduction [1] [6]An incredible year for Llama [7]Llama 3 models
  • Change · Modality Text-only, without the image input of the Llama 3.2 Vision models.Compared with Llama 3.2 Primary source[1]Model Information table, Input modalities
  • Change · Efficiency Meta says it offers performance similar to Llama 3.1 405B at a fraction of the serving cost.Compared with Llama 3.1 Primary source[6]An incredible year for Llama
  • Change · Tool use Uses the Llama 3.2 function-calling format, which Meta describes as more flexible than the Llama 3.1 format.Compared with Llama 3.1 Primary source[2]
  • Change · Availability Released only as an instruction-tuned model, without a pretrained version.Compared with Llama 3.1 Primary source[2]Introduction
  • Input text Primary source[1]Model Information table, Input modalities
  • Output text, code Primary source[1]Model Information table, Output modalities
  • Open weights YesApplies to Llama 3.3 70B, under the custom Llama 3.3 Community License after accepting the license. Primary source[3] [2]Download the Model
  • Context window 128k tokens Primary source[1]Model Information table, Context length
  • Access Open-weights download, APIAPI access through Meta's Llama API, which listed Llama-3.3-70B-Instruct in 2025 (archived page). Primary source[3] [8]Llama-3.3-70B-Instruct

Lineage

Predecessors

Successors

Based on

Not derived from another model.

Variants and derived

Siblings

None recorded.

All ancestors

All descendants

Variants

Llama 3.3 70B

Same dates as Llama 3.3.

  • Variant Inline variant in this record.Released only as an instruction-tuned model; Meta provides no pretrained version. Primary source[1]Model Information [2]Introduction [3] [7]Llama 3.3, Variations

Also known as: Llama-3.3-70B-Instruct, Llama 3.3 70B Instruct

Differs in:

  • Parameters: 70B Evidence not assessed [1]Model Information table, Params

Llama 3.3 8B

  • Announced Primary source[4]Introducing Llama API in preview
  • Released Date unknownNo release date found. From 2025-04-29 Meta offered fine-tuning of the model in the limited free preview of the Llama API. Evidence incompleteNo source
  • Variant Inline variant in this record.The archived Llama API documentation described it as a lightweight, fast variant of Llama 3.3 70B with a 128k-token context window and text input and output. Primary source[4]Introducing Llama API in preview [8]Llama-3.3-8B-Instruct

Status: Status unknown Evidence not assessed No source

Also known as: Llama-3.3-8B-Instruct

Differs in:

  • Open weights: Not recorded Evidence not assessed No source
  • Context window: 128k tokens tokens Evidence not assessed [8]Llama-3.3-8B-Instruct, Context window
  • Access: API Evidence not assessed [4]Introducing Llama API in preview [8]Llama-3.3-8B-Instruct

Related AI Radar coverage

AI Radar coverage starts in June 2026; no coverage linked yet.

All model releases from Meta on AI Radar →