Compare / DeepSeek V4 Pro 0423 vs Llama 3.3 70B Instruct
DeepSeek V4 Pro 0423 vs Llama 3.3 70B Instruct
Which is cheaper, which is more capable, and which does better at each kind of work, from 17 results on 3 sources that measured both models.
DeepSeek V4 Pro 0423 or Llama 3.3 70B Instruct? The short answer
- Llama 3.3 70B Instruct is cheaper: $0.64 against $1.98 per million tokens, 3.1 times less.
- DeepSeek V4 Pro 0423 scores higher on the Artificial Analysis Intelligence Index: 36.0 against 7.7.
- DeepSeek V4 Pro 0423 writes faster: 115 tokens a second against 80.
- On our business benchmarks, DeepSeek V4 Pro 0423 ranks higher for agents and tool use, reasoning and knowledge and writing.
At a glance
| DeepSeek V4 Pro 0423 | Llama 3.3 70B Instruct | |
|---|---|---|
| Price per million tokens | $1.32 in · $3.96 out | $0.59 in · $0.79 out |
| Blended (3 in : 1 out) | $1.98 | $0.64 |
| Cheapest host, blended | $0.26 (StreamLake, FP8) | $0.16 (DeepInfra, FP8) |
| Intelligence Index | 36.0 | 7.7 |
| Spring Prompt overall | 31 | – |
| Output speed, tokens a second | 115 | 80 |
| Context window | 1,048,576 tokens | 131,072 tokens |
| Weights | Open | Open |
| Developer based in | China | the US |
Prices are the developer's list price where we have read it, otherwise the typical host on OpenRouter; see every model's price. Intelligence and speed from Artificial Analysis.
Which is better for what
Each model's rank among models available today, for each kind of work, from our own benchmarks and licensed sources. The better rank is in green.
| Use case | DeepSeek V4 Pro 0423 | Llama 3.3 70B Instruct |
|---|---|---|
| Decks from an analysisTurning a finished analysis into a deck you could present as it is | 17 of 18 | not measured |
| Marketing planningPlanning a year of ad spend without overspending | 13 of 18 | not measured |
| Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls | 77 of 161 | 127 of 161 |
| Reasoning and knowledgeHard questions across science, maths and general knowledge | 28 of 195 | 174 of 195 |
| WritingWhat people prefer in blind comparisons, and judged writing quality | 29 of 187 | 165 of 187 |
| Sticking to the factsSummarising without inventing things, and factual answers | 55 of 153 | not measured |
| Speed under pressureGood decisions against a real clock (fast chess) | 22 of 22 | not measured |
Every shared result
| Benchmark | Metric | DeepSeek V4 Pro 0423 | Llama 3.3 70B Instruct |
|---|---|---|---|
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,460 | 1,304 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,446 | 1,285 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,480 | 1,298 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,452 | 1,293 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,458 | 1,318 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,451 | 1,292 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 68.8 | 11.9 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 36.0 | 7.70 |
| Artificial Analysisreported | GPQA Diamond% of questions | 92.8% | 49.8% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 41.0% | 3.6% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 80.3% | 15.7% |
| Artificial Analysisreported | Output speedtokens per second | 115 | 80.5 |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 78.7% | 4.9% |
| Artificial Analysisreported | Time to first answer tokenseconds | 0.91 | 0.64 |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 34.0% | 10.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.36 | 0.37 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 68.4 | 26.2 |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.
Quick answers
Is DeepSeek V4 Pro 0423 better than Llama 3.3 70B Instruct?
It depends on the task. Among models available today, on our benchmarks and licensed sources, DeepSeek V4 Pro 0423 ranks higher for agents and tool use, reasoning and knowledge and writing.
Which is cheaper, DeepSeek V4 Pro 0423 or Llama 3.3 70B Instruct?
Llama 3.3 70B Instruct costs $0.64 per million tokens (three input to one output), against $1.98 for DeepSeek V4 Pro 0423.
Which is faster, DeepSeek V4 Pro 0423 or Llama 3.3 70B Instruct?
DeepSeek V4 Pro 0423 writes about 115 tokens a second on its usual API, against 80 for Llama 3.3 70B Instruct (Artificial Analysis).
Run DeepSeek V4 Pro 0423 and Llama 3.3 70B Instruct on your own prompt
Benchmarks aren't your data. Try both side by side in the playground, with the cost of every answer.