

Llama 3.3 70B
#43 v Open-source jazykové modelyMeta · v3.3 · 70b · 10× · naposledy 10. 11. 2026
Llama 3.3 70B is an open-source, multilingual Large Language Model with 70 billion parameters developed by Meta. It is a pretrained and instruction-tuned text model (text input/output) that, according to Meta, achieves comparable performance to the significantly larger Llama 3.1 405B while requiring substantially less computational resources. The model employs an optimized Transformer architecture with Grouped-Query Attention, supports a context window of 128,000 Token, and is provided under the Llama 3.3 Community License. It is not multimodal but is designed for pure text tasks such as dialogue, coding, and multilingual applications.
Vlastnosti
| Key Benchmark (%) | MMLU (0-shot, CoT): 86.0% |
| Context Window (Tokens) | 128,000 tokens |
| License | Llama 3.3 Community License Agreement (custom commercial license, not an OSI-approved open-source license) |
| Multimodality | No, text-only model (no image or audio input) |
| Platform | Open model weights, available via Hugging Face, Meta's Llama site, Ollama, and cloud providers (AWS Bedrock, Oracle Cloud, NVIDIA NIM, etc.); no official first-party Meta API |
| Price per 1M Tokens | No first-party Meta pricing; e.g. Groq: $0.59 (input) / $0.79 (output) per 1M tokens |
| Release Date | December 6, 2024 |
Další produkty v této kategorii: Open-source jazykové modely
Zdroje (10)
Company Analysis: Meta
Data from the most recent quarter supports a strong operating backdrop (Q3 2026 revenue +26% YoY; ad impressions +10% YoY), but near-term earnings quality is clouded by large legal charges and a structurally higher AI/infrastructure cost trajectory. With the stock trading around ~21x trailing earnings (Nov 17, 2026) and with ongoing regulatory/legal headline risk (including a reported Italian privacy fine in mid-November 2026), the risk-adjusted setup is balanced: attractive fundamentals, but limited tolerance for negative surprises. A HOLD is warranted pending clearer evidence that margin resilience can be maintained alongside AI scaling and that legal/regulatory costs do not recur at similar magnitude.
Summary
Meta Platforms (META) operates a two-segment model: (1) Family of Apps (FoA)—Facebook, Instagram, WhatsApp, Messenger—monetized primarily via performance advertising; and (2) Reality Labs (RL)—AR/VR hardware, software, and AI-enabled wearables, currently loss-making but positioned as a longer-duration option on new computing platforms. Core competencies are global-scale consumer product distribution, ad targeting/measurement, and increasingly AI infrastructure and model deployment that improves ranking, recommendations, and ad performance. Market position remains top-tier in global digital advertising, with competitive advantages in user scale, engagement, and a broad advertiser base. In the last ~90 days, the most material company-specific datapoint is Meta’s Q3 2026 results and commentary: revenue of $64.60B (+26% YoY) with ad impressions up 10% YoY across FoA, indicating continued demand and effective monetization improvements. However, reported profitability was distorted by a large legal-related charge: Q3 2026 expenses included a $9.89B charge tied to an August 2026 multistate attorneys general agreement; Meta also highlighted rising infrastructure expenses and third-party AI token costs. Excluding the legal charge, Meta indicated operating income would have increased YoY, supporting the view that underlying operating leverage remains intact despite elevated AI-related cost growth. Reality Labs revenue was $443M (down 6% YoY) due to lower Quest headset sales, partially offset by growth in AI glasses revenue, reinforcing that RL is still early and volatile. Valuation/metrics (as of Nov 17, 2026): META last price is $626.41; using EUR/USD ~1.0997 (USD per EUR), this implies ~€569.62 per share. Trailing P/E is ~21.36x and TTM EPS ~29.32. Market cap is ~US$1.61T. Near-term investor focus is likely to remain on (a) durability of ad growth, (b) the pace of AI infrastructure spend and its margin impact, and (c) regulatory/legal overhangs (e.g., a recent Italian privacy fine reported in mid-November 2026). Outlook (short- to medium-term): data from the latest quarter supports strong top-line momentum and improving AI-driven monetization, but expense volatility (legal items) and structurally higher AI/infrastructure costs can create earnings variability. The base case is continued revenue growth into 2027 supported by AI improvements, with margins dependent on the slope of capex/opex and the frequency of regulatory/legal costs.
Key Takeaways
- Meta’s core earnings engine remains FoA advertising; Q3 2026 revenue grew 26% YoY to $64.60B with ad impressions up 10% YoY, indicating strong demand and monetization execution. (Q3 2026 release)
- Reported Q3 2026 profitability was heavily impacted by a $9.89B legal-related charge tied to an August 2026 multistate AG agreement; excluding this, Meta stated operating income would have increased YoY. (Q3 2026 release/call transcript)
- AI is a dual driver: it is improving ranking/recommendations and ad performance, but also increasing cost of revenue and infrastructure expenses (data centers, cloud, and third-party AI token costs). (SEC filing/Q3 materials)
- Reality Labs remains a drag and is still proving product-market fit: Q3 RL revenue was $443M (down 6% YoY) with weaker Quest sales, partly offset by AI glasses growth. (earnings call transcript)
- At ~21.36x trailing P/E (Nov 17, 2026), valuation is not “low,” so the investment case depends on sustained revenue growth and disciplined cost/capex execution rather than multiple expansion alone. (market data)
Action Ideas
Accumulate on the view that FoA ad growth is being structurally supported by AI-driven ranking and measurement improvements. Q3 2026 showed 26% YoY revenue growth and 10% YoY ad impression growth, while management commentary indicates AI investments are benefiting the core business. The large Q3 legal charge is material but non-recurring in nature; normalized operating performance (as described by the company) was stronger than GAAP results suggest. Entry should be sized with awareness that AI infrastructure costs can pressure margins even in a strong revenue environment.
Horizon: 12 mo.
Maintain exposure if already owned, emphasizing that the stock’s trailing valuation (~21x P/E) implies the market already prices in continued solid growth. The Q3 2026 print supports strong fundamentals, but the path of expenses (legal items plus AI-related opex/capex) can create quarter-to-quarter volatility. A hold stance is appropriate while monitoring whether revenue growth remains strong through 2027 and whether cost growth moderates as AI infrastructure scales.
Horizon: 6 mo.
Reduce exposure if portfolio constraints require lower volatility or if the investment mandate cannot tolerate regulatory/legal headline risk. Recent enforcement actions (e.g., Italy privacy fine reported Nov 2026) highlight that compliance costs and product changes can be recurring. If AI-related costs continue to rise (infrastructure, cloud, token costs) without commensurate margin resilience, the risk-adjusted return profile may deteriorate even if revenue remains strong.
Horizon: 3 mo.
Contrarian Insights
- • Consensus often frames Meta’s AI spend primarily as a growth catalyst; a more conservative read is that third-party AI token costs and infrastructure expenses can become a structurally higher cost base, limiting margin expansion even with strong revenue growth. (SEC filing commentary on cost drivers)
- • Reality Labs is widely viewed as a near-term drag; contrarian upside is that AI glasses revenue growth (noted in Q3 commentary) could indicate earlier-than-expected traction in wearables, potentially improving the narrative even before RL becomes profitable. (earnings call transcript)
Sources (7)
- https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Reports-Third-Quarter-2026-Results/default.aspx
- https://www.sec.gov/Archives/edgar/data/1326801/000162828026066773/meta-20260930.htm
- https://s21.q4cdn.com/399680738/files/content_files/META-Q3-2026-Earnings-Call-Transcript.pdf
- https://apnews.com/article/9b3073c52464d5cd0e8facbc57285194
- https://finance.yahoo.com/quote/META/analysis/
- https://stockanalysis.com/stocks/meta/forecast/
- https://wise.com/us/currency-converter/eur-to-usd-rate/history/17-11-2026