ResearchLLM Verifications Leaderboard

Data as of · 78 models, 80 questions · reviewed by Suhas Bhairav · methodology

MetaOpen weightsReleased 6 Dec 2024

Llama 3.3 70B Instruct benchmarks, speed and API pricing

Reviewed by Suhas Bhairav · Tested

Compare

Available through OpenRouter on

  • DeepInfra$0.1 in · $0.32 out
  • Novita$0.135 in · $0.4 out
  • Parasail$0.22 in · $0.5 out
  • AkashML$0.2 in · $0.52 out
  • CoreWeave$0.71 in · $0.71 out
  • Google$0.72 in · $0.72 out
  • Groq$0.59 in · $0.79 out
  • SambaNova$0.45 in · $0.9 out
  • Cloudflare$0.293 in · $2.253 out
Intelligence42.5% correct#52 of 72
Speed42tok/s#63 of 72
Cost$0.00053per question#18 of 72List $0.1 in · $0.32 out per 1M
Verbosity855tokens per question#28 of 72

Summary

  • Answers 34 of 80 verification questions correctly (42.5%), placing #52 of 72 models tested on the current suite.
  • Generates 42 output tokens per second, faster than 13% of the models the leaderboard measures.
  • Starts responding after 0.7 s (median time to first token).
  • Costs $0.00053 per question as billed, #9 cheapest of 29 models priced $0.15 to $1 per 1M tokens.

Specifications

Reasoning
No
Input
text
Output
text
Context window
131k tokens
Knowledge cutoff
31 Dec 2023
OpenRouter ID
meta-llama/llama-3.3-70b-instruct
Last tested
26 Oct 2026

Peer group: 31 models with a blended list price $0.15 to $1 per 1M tokens (3 input : 1 output). Charts below compare Llama 3.3 70B Instruct with all models; cost and speed charts focus on this peer group.

Intelligence

Verification score: the share of 80 questions from open datasets answered correctly. See the benchmark datasets.

Score vs. cost per question

Llama 3.3 70B Instruct circled; dashed line joins models no cheaper model beats

Most attractive quadrant0255075100$0.00001$0.0001$0.001$0.01$0.10Cost per question (USD, log) →Verification score (%) →Qwen3.8 2.4T A95BCommand AReka Flash 3Gemma 4 26B A4BGemma 4 31BQwen3.5 397B A17BHermes 4 405BMuse WonderGrok 4.7o3gpt-oss-20bGranite 4.0 MicroNova Micro 1.0Mercury 2.5Phi 4Granite 4.2 8BHy3GLM 5.3 FlashSolar Pro 4Trinity Large ThinkingMiniMax M3Mistral Large 4Claude Opus 5.5GPT-6.1 SolGemini 4 ArgonGPT-5.4GPT-6 Lunagpt-oss-120b

Score vs. end-to-end response time

Llama 3.3 70B Instruct circled · accurate and quick sits top left

Most attractive quadrant02550751000 s25 s50 s75 s100 sEnd-to-end response time (s) →Verification score (%) →Qwen3.8 2.4T A95BCommand AReka Flash 3Llama 4 ScoutGemma 4 26B A4BLing 3.1 FlashHermes 4 405BMuse WonderGranite 4.0 MicroSolar Pro 4DeepSeek V4 Pro 0813Mistral Large 4Claude Opus 5.5Llama 4 MaverickKimi K3.1 Pro

Score vs. tokens per question

Llama 3.3 70B Instruct circled · output tokens, reasoning included

Most attractive quadrant02550751001001,00010,000Output tokens per question (log) →Verification score (%) →Command AReka Flash 3Llama 4 ScoutQwen3.7 PlusGemma 4 26B A4BGemma 4 31BQwen3.5 397B A17BHermes 4 405BClaude Fable 5.1GPT-6 AstraMuse WonderGrok 4.7o3gpt-oss-20bGranite 4.0 MicroNova Micro 1.0Mercury 2.5Granite 4.2 8BMinistral 3 14B 2512Hy3GLM 5.3 FlashSolar Pro 4Qwen3.6 35B A3BInkling SmallNemotron 3 UltraClaude Opus 5.5Gemini 4 ArgonGPT-5.4GPT-6 SolMiMo-V2.6-Pro

Verification score

All models tested on the current suite

  1. 1Claude Opus 5.585.0
  2. 2Gemini 4 Argon83.8
  3. 3GPT-5.3-Codex82.5
  4. 3Gemini 3.8 Flash82.5
  5. 5Claude Fable 5.181.2
  6. 6Grok 4.780.0
  7. 7Muse Wonder78.8
  8. 7Claude Sonnet 5.578.8
  9. 9Gemini 3.1 Pro Preview77.5
  10. 9Muse Spark Open77.5
  11. 52Llama 3.3 70B Instruct42.5

61 more models · see the full leaderboard

6 of 78 tested models have no verification score

Cost

Average cost of one question, split into input, reasoning and answer tokens at each model's list price. Reasoning tokens are billed as output.

Cost per question, by token type

Peer group: blended price $0.15 to $1

  1. 1Llama 4 Scout$0.00025
  2. 2GPT-6 Luna$0.00025
  3. 3Ministral 3 8B 2512$0.00027
  4. 4Llama 3.3 70B Instruct$0.00031
  5. 5Ministral 3 14B 2512$0.00031
  6. 6GLM 5.3 Flash$0.00033
  7. 7gpt-oss-120b$0.00038
  8. 8Inkling Small$0.00047
  9. 9Claude Haiku 5.5$0.00048
  10. 10MiMo-V2.6-Flash$0.00056
  11. 11Llama 4 Maverick$0.00057
  12. 12GLM 5.3$0.00076
  13. 13DeepSeek V4.1 Flash 1017$0.00084
  14. 14MiniMax M3.1 Flash$0.0009
  15. 15DeepSeek V4 Flash Vision Exp$0.00092
  16. 16Hy3$0.0013
  17. 17Gemma 4 31B$0.0014
  18. 18Nemotron 3.5 Super VL$0.0014
  19. 19Qwen3 Coder Next$0.0017
  20. 20Solar Pro 4$0.0018
  21. 21Mistral Small 4$0.0019
  22. 22MiMo-V2.6-Pro$0.002
  23. 23Trinity Large Thinking$0.0025
  24. 24MiniMax M3$0.0027
  25. 25LongCat 2.0$0.0032
  26. 26Qwen3.7 Plus$0.0033
  27. 27Gemini 3.5 Flash Lite$0.0034
  28. 28Qwen3.6 35B A3B$0.0034
  29. 29Qwen3.8 27B$0.0049
Input Reasoning Answer

Cost per question, as billed

All models · lower is better

  1. 1Granite 4.0 Micro$0.000051
  2. 2gpt-oss-20b$0.000066
  3. 3Granite 4.2 8B$0.000077
  4. 4Nova Micro 1.0$0.000087
  5. 5Mercury 2.5$0.00011
  6. 6Phi 4$0.00012
  7. 7gpt-oss-120b$0.00013
  8. 8GLM 5.3 Flash$0.00015
  9. 9Ministral 3 3B 2512$0.00019
  10. 10GPT-6 Luna$0.00025
  11. 18Llama 3.3 70B Instruct$0.00053

61 more models · see the full leaderboard

6 of 78 tested models have no cost per question
  • LFM2.5-2.6B: no verification score (80 of 80 questions failed on the host's side)
  • Ling 3.0 Flash VL: no verification score (not yet run on the current question suite)
  • Qwen3.5-122B-A10B: no verification score (not yet run on the current question suite)
  • Qwen3 Next 80B A3B Thinking: no verification score (not yet run on the current question suite)
  • Nova Premier 1.0: no verification score (not yet run on the current question suite)
  • GPT-5.5 Pro: no verification score (not yet run on the current question suite)

Speed and latency

Median of 3 streamed runs of the same prompt, measured 25 Oct 2026.

Output speed

Peer group ($0.15 to $1) · higher is better

  1. 1Nemotron 3.5 Super VL715
  2. 2Gemini 3.5 Flash Lite409
  3. 3Trinity Large Thinking329
  4. 4Inkling Small233
  5. 5Claude Haiku 5.5192
  6. 6Mistral Small 4170
  7. 7DeepSeek V4 Flash Vision Exp170
  8. 8GLM 5.3154
  9. 9GPT-6 Luna137
  10. 10Qwen3.6 35B A3B135
  11. 11MiniMax M3.1 Flash131
  12. 12Ministral 3 8B 2512124
  13. 26Llama 3.3 70B Instruct42

16 more models · see the full leaderboard

2 of 31 tested models have no output speed

Time to first token

Peer group ($0.15 to $1) · lower is better

  1. 1Ministral 3 8B 25120.3 s
  2. 2MiMo-V2.6-Flash0.4 s
  3. 3Qwen3.6 35B A3B0.4 s
  4. 4Qwen3 Coder Next0.4 s
  5. 5Mistral Small 40.5 s
  6. 6GLM 5.3 Flash0.5 s
  7. 7Inkling Small0.5 s
  8. 8MiniMax M30.5 s
  9. 9MiniMax M3.1 Flash0.5 s
  10. 10Qwen3.8 27B0.5 s
  11. 11gpt-oss-120b0.5 s
  12. 12Trinity Large Thinking0.6 s
  13. 20Llama 3.3 70B Instruct0.7 s

16 more models · see the full leaderboard

2 of 31 tested models have no time to first token

End-to-end response time

Peer group ($0.15 to $1) · lower is better

  1. 1Nemotron 3.5 Super VL1.5 s
  2. 2Trinity Large Thinking2.7 s
  3. 3Gemini 3.5 Flash Lite2.8 s
  4. 4DeepSeek V4 Flash Vision Exp3.2 s
  5. 5GLM 5.33.4 s
  6. 6Ministral 3 8B 25123.5 s
  7. 7Inkling Small3.6 s
  8. 8Claude Haiku 5.53.9 s
  9. 9MiniMax M3.1 Flash4.2 s
  10. 10Qwen3 Coder Next4.2 s
  11. 11MiniMax M34.2 s
  12. 12GPT-6 Luna4.3 s
  13. 21Llama 3.3 70B Instruct8.8 s

16 more models · see the full leaderboard

2 of 31 tested models have no end-to-end response time

Latency vs. output speed

Llama 3.3 70B Instruct circled · quick to start and fast sits top left

Most attractive quadrant02004006008000.0 s5.0 s10.0 s15.0 s20.0 sTime to first token (s) →Output speed (tok/s) →Nemotron 3.5 Super VLQwen3.5 397B A17BMuse WonderMinistral 3 3B 2512Trinity Large ThinkingGemini 3.5 Flash LiteGemini 3.1 Pro PreviewGemini 3.8 FlashGemini 4 Argon

Results by category

Questions answered correctly in each category

Science11/30
Engineering2/15
Medicine6/11
Economics5/9
Law1/6
Humanities & social science9/9

More from Meta

Meta#7Muse Wonder78.8% verification score175 tok/s16.1 s first token$0.021 / questionMeta#9Muse Spark Open77.5% verification score220 tok/s0.5 s first token$0.0045 / questionMeta#40Llama 4 Maverick59.0% verification score13 tok/s0.7 s first token$0.00059 / question

All Meta models →

Compare Llama 3.3 70B Instruct

Head to head with its closest rivals, other Meta models and the top models overall.

Hosts and prices

Every host serving this model through OpenRouter, from OpenRouter's catalogue

HostInput / 1MOutput / 1MContextMax outputQuantization
DeepInfra$0.1$0.32131k16kfp8
Novita$0.135$0.412k11kbf16
Parasail$0.22$0.5131k16kfp8
AkashML$0.2$0.52131k128kfp8
CoreWeave$0.71$0.71128k115kfp16
Google$0.72$0.72128k8k—
Groq$0.59$0.79131k33k—
SambaNova$0.45$0.9131k3k—
Cloudflare$0.293$2.25324k22kfp8

Llama 3.3 70B Instruct: frequently asked questions

How accurate is Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct answered 34 of 80 questions correctly (42.5%) in the 2026-10-27 run, ranking #52 of 78 models.
How fast is Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct generated a median of 42 output tokens per second in the 2026-10-27 run, ranking #63 of 78. Median time to first token was 0.7 s.
What is the latency of Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct had a median time to first token of 0.7 s in the 2026-10-27 run, ranking #33 of 78 models.
How much does Llama 3.3 70B Instruct cost?
Llama 3.3 70B Instruct cost $0.00053 per question on average as billed in the 2026-10-27 run, reasoning tokens included.
What is Llama 3.3 70B Instruct API pricing?
Llama 3.3 70B Instruct is listed at $0.1 per million input tokens and $0.32 per million output tokens on OpenRouter.
What is the context window of Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct accepts up to 131,072 tokens of context, according to OpenRouter's model catalogue.
Is Llama 3.3 70B Instruct open source?
Yes. Llama 3.3 70B Instruct is an open-weights model: Meta publishes the weights, so you can run it yourself as well as through an API.
Who makes Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct is made by Meta. The LLM Verifications Leaderboard tested it through OpenRouter using the model ID meta-llama/llama-3.3-70b-instruct.
Where can I use Llama 3.3 70B Instruct via API?
Through OpenRouter, Llama 3.3 70B Instruct is served by 6 hosts: AkashML, Cloudflare, DeepInfra, Groq, Novita, Parasail.