Llama 3.1 8B Instruct
Llama 3.1 8B InstructCompare inon
Benchmarks of Llama 3.1 8B Instruct20
BenchmarkByScoreRank
Math1
Writing15
Legacy · saturated or no longer reported4
No benchmark matches
Rankingconfiguration unstated?
Alternatives?
ModelScoreCostPer taskWhat changes
BetterGPT-6 Astraunstated×3+79,900%$10 / $50Monitoring can return HTTP 403+2
OpenGemma 4 12Bunstated+81%+500%$0.10 / $0.30Memory (4-bit): 4 GB → 6 GB
BetterClaude Fable 5.1unstated×3+79,900%$10 / $50Replace forced tool_choice+5BetterClaude Opus 5unstated×3+39,900%$5 / $25BetterGPT-6 Solunstated×3+15,900%$2 / $10ModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-6 Astraunstated×3+79,900%$10 / $50Monitoring can return HTTP 403+2Claude Fable 5.1unstated×3+79,900%$10 / $50Replace forced tool_choice+5Claude Opus 5unstated×3+39,900%$5 / $25GPT-6 Solunstated×3+15,900%$2 / $10
Kimi K3unstatedOpen×2.9+23,900%$3 / $15Memory (4-bit): 4 GB → 1,668 GB+4ModelScoreCostPer taskWhat changes
Gemma 4 12BunstatedOpen+81%+500%$0.10 / $0.30Memory (4-bit): 4 GB → 6 GB
NVIDIA Nemotron 3.5 Lightning 30B A3B NVFP4Open+80%——Memory (4-bit): 4 GB → 11 GBGemma 3 12BunstatedOpen+58%+150%$0.05 / $0.10Memory (4-bit): 4 GB → 6.1 GBGemma 3 4BunstatedOpen+50%+70%$0.04 / $0.05
Mistral NemounstatedOpen+24%+500%$0.15 / $0.15Memory (4-bit): 4 GB → 6.1 GBModelScoreCostPer taskWhat changes
Muse Spark 1.1unstated×2.7—$1.25 / $4.25Only n=1 · Meta / Chat Completions+1Muse Spark 1.3unstated×2.7—$1.25 / $4.25Only auto tool_choice is supported+5Muse Spark 1.2unstated×2.6—$1.25 / $4.25Only auto tool_choice is supported+1Muse Glimmer 30BunstatedOpen×2.5—$0.30 / $1.10Memory (4-bit): 4 GB → 18 GBLlama 4 MaverickunstatedOpen+21%—$0.19 / $0.65Memory (4-bit): 4 GB → 241 GBLlama 3.1 70B InstructunstatedOpen+10%—$0.40 / $0.40Memory (4-bit): 4 GB → 42 GBLlama 4 ScoutunstatedOpen+10%—$0.10 / $0.30Memory (4-bit): 4 GB → 65 GBLlama 3.2 3B InstructunstatedOpen-17%—$0.02 / $0.02All Meta models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · compressed✓ 16 GB · 8-bit✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceLlama 3.1 licencecommercial use with conditions
Benchmarks
all categoriesChat & Knowledge
Arena Text · overall1,211#182
Arena Text · Chinese1,150.8#165
Arena Text · English1,217.5#168
Arena Text · French1,177.5#137
Arena Text · German1,144.3#141
Arena Text · instruction following1,158.8#170
Arena Text · Japanese1,060.9#127
Arena Text · Korean1,052.8#136
Arena Text · non-English1,148#169
Arena Text · Russian1,158.3#169
Arena Text · Spanish1,169.2#144
MATH Level 522.9#28
Arena Text · overall · full cohort1,186.5#171
GPQA Diamond · Epoch27#124
MMLU-Pro · official board44.3#21
OTIS Mock AIME 2024 20251.7#122
Agents & Tool Use
BFCL v425.8#30
Writing & Language
Arena Text · creative writing1,177#182
Arena Text · creative writing · full cohort1,154.2#169
FAQ
How much does Llama 3.1 8B Instruct cost?›
$0.02 per 1M input tokens and $0.04 per 1M output tokens (cheapest host: DeepInfra).
Is Llama 3.1 8B Instruct free to use?›
There is no free API; the cheapest paid price is $0.02 per 1M input tokens. The weights are open, so you can run it yourself.
How good is Llama 3.1 8B Instruct, and what is it best at?›
#182 of 196 in Chat & Knowledge, #30 of 39 in Agents & Tool Use, #182 of 195 in Writing & Language.
Is Llama 3.1 8B Instruct better than Mistral Small 3.1 24B?›
On EQ-Bench Creative Writing v3, Llama 3.1 8B Instruct scores 713 vs 760.9 — lower.
How can I use Llama 3.1 8B Instruct?›
Through the API of Cloudflare, CoreWeave, DeepInfra, Groq, Novita.
What is the context window of Llama 3.1 8B Instruct?›
131K tokens, with up to 118K output tokens.
Is Llama 3.1 8B Instruct open source?›
Yes, the weights are open (llama3.1 licence).
Can I run Llama 3.1 8B Instruct locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Llama 3.1 8B Instruct read images?›
No, it accepts text only.
What is Llama 3.1 8B Instruct?›
Llama 3.1 8B Instruct is an open-weights language model from Meta.
What are the best alternatives to Llama 3.1 8B Instruct?›
GPT-6 Astra (better), Gemma 4 12B (open weights).