Llama 3.3 70B Instruct
Meta · released Nov 26, 2024 · meta-llama/Llama-3.3-70B-Instruct
- Type
- Open weightsllama3.3
- Params
- 70.6B
- Context
- 131K
about 98K words of context · download allowed, licence restricts use
Our take
Written Aug 3, 2026Llama 3.3 is a 70.6-billion-parameter text model from Meta with a 131,072-token request limit and a licence that permits commercial use with specific conditions. It is a strong instruction-follower with wide hosting choice, though its reasoning scores sit below what its size might suggest.
Pick this for high-volume text work where the licence terms are acceptable, or for instruction-following workflows. Use it for long-document processing up to 131,072 tokens, or where low latency matters — some hosts deliver 68 tokens per second. Skip it if you need multimodal input, a fully permissive licence, or graduate-level reasoning.
The case for it
- Exceptional instruction-following capability: IFEval 90% stands well above its other benchmark scores.
- Wide provider choice with meaningful price competition at the low end.
- Strong coding performance in live human evaluation: Arena Coding Elo 1345.6, the highest of its six Arena sub-scores.
- 131,072-token request limit for a fully downloadable model of this parameter class, with no mixture-of-experts gating.
The case against it
- Weak graduate-level reasoning and broad knowledge for its scale: GPQA Diamond 10.5% and MMLU-Pro 48.1% are low enough that many smaller or similarly sized models likely outperform it.
- No measured throughput on three of ten offers, including one Novita endpoint.
- Licence carries usage restrictions that Apache or MIT licences do not: attribution required, specific domains prohibited, and compliance obligations above 700 million users.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)116th of 143 · 1318.4
CodingWriting and fixing code on its own
Arena Coding122nd of 143 · 1345.6
Arena Coding is the only board that has scored it for this.
AgenticPlanning, calling tools, staying on task
Nobody we watch has scored Llama 3.3 70B Instruct for this. We would take the rating from Arena Agent (IPS).
WritingWe do not rate this
Two boards come close and neither tests writing: Arena Creative Writing asks people which of two replies they prefer, and LiveBench Language tests whether a model understood a passage. So we show where Llama 3.3 70B Instruct placed and give it no mark out of five.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.
Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Max (32-core GPU) · 64 GB
Borderline fit on an estimated size. It leaves 0.6 GB spare on a size we calculated rather than measured, and a 10% error either way would change the answer.
Memory use by level
Against a 24 GB card.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
Or rent it from someone else
Cheapest of 17 live listings. Picked at the widest standard context we hold, within one quantisation slice, so the numbers beside it are a price one host actually charges.
- per 1M tokens
- $0.10 in / $0.32 out
- Context served
- 131K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| DeepInfraturbo tierfp8 | $0.10 / $0.32 | 131K | 17 tok/s | No | No | Unknown |
| DeepInfrafp8 | $0.10 / $0.32 | 131K | 7 tok/s | No | No | Unknown |
| OpenRouter | $0.10 / $0.32 | 131K | not measured | Unknown | Unknown | Unknown |
| AkashMLfp8 | $0.13 / $0.40 | 131K | 21 tok/s | No | No | Confirmed |
| Novita AIbf16 | $0.14 / $0.40 | 6K | 24 tok/s | No | No | Confirmed |
| Novita AI | $0.14 / $0.40 | 6K | not measured | Unknown | Unknown | Unknown |
| Nebius AI Studiofp8 | $0.13 / $0.40 | 131K | 16 tok/s | No | No | Confirmed |
| Parasailfp8 | $0.22 / $0.50 | 131K | 35 tok/s | No | No | Confirmed |
| CoreWeavefp16 | $0.71 / $0.71 | 128K | 71 tok/s | No | No | Confirmed |
| Google Vertex AIus-central1 | $0.72 / $0.72 | 128K | 53 tok/s | No | No | Confirmed |
| Google Vertex AI | $0.72 / $0.72 | 128K | 51 tok/s | No | No | Confirmed |
| Crusoebf16 | $0.25 / $0.75 | 131K | 60 tok/s | No | No | Confirmed |
| Groq | $0.59 / $0.79 | 131K | 179 tok/s | No | No | Confirmed |
| SambaNovabf16 | $0.45 / $0.90 | 131K | 103 tok/s | Unknown | Unknown | Unknown |
| Together AIfp8 | $1.04 / $1.04 | 131K | 42 tok/s | No | No | Confirmed |
| SambaNova | $0.60 / $1.20 | 131K | not measured | Unknown | Unknown | Unknown |
| Cloudflare Workers AIfp8 | $0.29 / $2.25 | 24K | 43 tok/s | No | Yesunknown period | Unknown |
DeepInfra
Novita AI
Nebius AI Studio
Google Vertex AI
Groq
Together AI
SambaNova
Cloudflare Workers AIAcross the 17 listings we hold: 13 say they do not train on prompts, 0 say they do and 4 do not say. 10 appear in the zero-retention registry we check; the rest are unknown to us rather than confirmed either way.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
- ✓
- Supported
- ✗
- Not supported
- Not published
- host gave no parameter list
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| DeepInfraturbo · fp8 | ✓ | ✓ | ✗ |
| DeepInfrafp8 | ✓ | ✓ | ✗ |
| OpenRouter | ✓ | ✓ | ✓ |
| AkashMLfp8 | ✓ | ✓ | ✓ |
| Novita AIbf16 | ✓ | ✓ | ✗ |
| Novita AI | |||
| Nebius AI Studiofp8 | ✓ | ✓ | ✓ |
| Parasailfp8 | ✗ | ✓ | ✓ |
| CoreWeavefp16 | ✓ | ✓ | ✓ |
| Google Vertex AIus-central1 | ✓ | ✓ | ✓ |
| Google Vertex AI | ✗ | ✓ | ✗ |
| Crusoebf16 | ✗ | ✓ | ✓ |
| Groq | ✓ | ✓ | ✗ |
| SambaNovabf16 | ✗ | ✓ | ✓ |
| Together AIfp8 | ✓ | ✓ | ✓ |
| SambaNova | |||
| Cloudflare Workers AIfp8 | ✗ | ✓ | ✗ |
Tool calling: 10 of 17 listings say yes, 5 say no, 2 publish no parameter list. JSON output: 15 of 17 listings say yes, 2 publish no parameter list. Strict schema: 9 of 17 listings say yes, 6 say no, 2 publish no parameter list.
Models people weigh against Llama 3.3 70B Instruct
When we formed this view
Dates behind this page
Prices last checked 14h ago
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- 2 of 17 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 4 of 17 listings do not say whether they train on prompts.
Licence and identifiers
What the licence allowsllama3.3, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
llama3.3
License tag "llama3.3" imported from Hugging Face; terms pending curation — review the original text before relying on it.
Identifiers
- Hugging Face
- meta-llama/Llama-3.3-70B-Instruct
- Modality record
- text->text
- Catalogue slug
- meta-llama-llama-3-3-70b-instruct