Llama 3.3 70B Instruct
Nov 2024
Llama 3.3 70B InstructCompare inon
Benchmarks of Llama 3.3 70B Instruct26
BenchmarkByScoreRank
Math1
Writing15
Legacy · saturated or no longer reported10
No benchmark matches
Rankingconfiguration unstated?
Alternatives?
ModelScoreCostPer taskWhat changes
CheaperGemma 3 27BunstatedOpenwins 59%-35%$0.08 / $0.16
BetterGPT-5 Minihighwins 55%+344%$0.053OpenGemma 4 31Bunstatedwins 68%+3%$0.10 / $0.34
CheaperQwen 3.5 Flashunstatedwins 58%-27%$0.065 / $0.26CheaperGemma 3 12BunstatedOpenwins 57%-60%$0.05 / $0.10ModelScoreCostPer taskWhat changes
Gemma 3 27BunstatedOpenwins 59%-35%$0.08 / $0.16Qwen 3.5 Flashunstatedwins 58%-27%$0.065 / $0.26Gemma 3 12BunstatedOpenwins 57%-60%$0.05 / $0.10
GLM 4.7 FlashunstatedOpenwins 53%-6%$0.061 / $0.40GPT OSS 120BunstatedOpensame-61%$0.03 / $0.15Memory (4-bit): 42 GB → 70 GBModelScoreCostPer taskWhat changes
GPT-5 Minihighwins 55%+344%$0.053GPT-6 Lunamaxwins 67%+29%$0.068Function calling only with none · Chat Completions+3GPT-5.6 Lunaxhighwins 67%+190%$0.085GPT-5.6 Terraxhighwins 70%+2,803%$0.63Prompts with >272K input tokensGPT-6.1 Solmaxwins 73%+2,481%$0.72Remove unsupported parameters+5ModelScoreCostPer taskWhat changes
Gemma 4 31BunstatedOpenwins 68%+3%$0.10 / $0.34
Mistral Medium 3.5unstatedOpenwins 66%+1,835%$1.50 / $7.50Memory (4-bit): 42 GB → 77 GBGemma 4 26B A4BunstatedOpenwins 66%+69%$0.15 / $0.60Qwen3.8 27BunstatedOpenwins 61%+517%$0.42 / $2.55max_tokens range · QwenCloud+4Qwen 3.5 122B A10BunstatedOpenwins 61%+610%$0.40 / $3.20Memory (4-bit): 42 GB → 75 GBModelScoreCostPer taskWhat changes
Muse Wondermaxwins 76%—$4.52Only auto tool_choice is supported+5Muse Sparkunstatedwins 74%—$1.25 / $4.25Only n=1 · Meta / Chat Completions+1Muse Spark 1.3maxwins 73%—$1.60Only auto tool_choice is supported+5Muse Spark 1.2xhighwins 72%—$0.97Only auto tool_choice is supported+1Muse Spark 1.1unstatedwins 71%—$1.25 / $4.25Only n=1 · Meta / Chat Completions+1Muse Glimmer 30BunstatedOpenwins 61%—$0.30 / $1.10Llama 4 MaverickunstatedOpenwins 53%—$0.19 / $0.65Memory (4-bit): 42 GB → 241 GBLlama 4 ScoutunstatedOpensame—$0.10 / $0.30Memory (4-bit): 42 GB → 65 GBAll Meta models →Run it yourself
estimateFits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✓ 48 GB GPU · compressed✓ 80 GB GPU · 8-bit
LicenceLlama 3.3 licencecommercial use with conditions
History
2 versionsBenchmarks
all categoriesChat & Knowledge
Arena Text · overall1,318#145
Arena Text · Chinese1,217#155
Arena Text · English1,303.3#146
Arena Text · French1,281.5#123
Arena Text · German1,250.7#124
Arena Text · instruction following1,241.7#153
Arena Text · Japanese1,150.2#117
Arena Text · Korean1,142.6#128
Arena Text · non-English1,236.3#154
Arena Text · Polish1,259.2#121
Arena Text · Russian1,251.7#149
Arena Text · Spanish1,269.5#132
MATH Level 541.6#22
Arena Text · overall · full cohort1,274.2#152
GPQA Diamond · Epoch47.4#107
OTIS Mock AIME 2024 20255.1#114
Coding
SWE-rebench · all · 2025-01-01 to 2025-09-0111.5#1
SWE-rebench · Go · 2025-01-01 to 2025-09-010#1
SWE-rebench · Java · 2025-01-01 to 2025-09-010#1
SWE-rebench · Rust · 2025-01-01 to 2025-09-010#1
SWE-rebench · Python · 2025-01-01 to 2025-09-0111.5#1
SWE-rebench · TypeScript · 2025-01-01 to 2025-09-010#1
Agents & Tool Use
BFCL v431.9#23
Writing & Language
Arena Text · creative writing1,285#139
Arena Text · creative writing · full cohort1,250.2#146
FAQ
How much does Llama 3.3 70B Instruct cost?›
$0.10 per 1M input tokens and $0.32 per 1M output tokens (cheapest host: DeepInfra).
Is Llama 3.3 70B Instruct free to use?›
There is no free API; the cheapest paid price is $0.10 per 1M input tokens. The weights are open, so you can run it yourself.
How good is Llama 3.3 70B Instruct, and what is it best at?›
#145 of 196 in Chat & Knowledge, #23 of 39 in Agents & Tool Use, #139 of 195 in Writing & Language.
Is Llama 3.3 70B Instruct better than Llama 4 Scout?›
On Arena Text · creative writing, Llama 3.3 70B Instruct scores 1,285 vs 1,289 — about the same.
How can I use Llama 3.3 70B Instruct?›
Through the API of AkashML, Cloudflare, CoreWeave, DeepInfra, Google and 4 more.
What is the context window of Llama 3.3 70B Instruct?›
128K tokens, with up to 4K output tokens.
When was Llama 3.3 70B Instruct released? Is there a newer version?›
Released 2024-11-26. It is the current version in its line.
Is Llama 3.3 70B Instruct open source?›
Yes, the weights are open (llama3.3 licence).
Can I run Llama 3.3 70B Instruct locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Llama 3.3 70B Instruct read images?›
No, it accepts text only.
What is Llama 3.3 70B Instruct?›
Llama 3.3 70B Instruct is an open-weights language model from Meta, released Nov 2024.
What are the best alternatives to Llama 3.3 70B Instruct?›
Gemma 3 27B (cheaper), GPT-5 Mini (better), Gemma 4 31B (open weights).