1. Anthropic’s Claude 3 Is Holding Its Own Among Competitors
Last week, Anthropic released Claude 3, its best performing suite of large language models (LLMs) to date. Claude 3 Opus outscores GPT-4 marginally—86.8% to 86.4%—on the MMLU (Massive Multi-task Language Understanding) benchmark, a broad-based undergraduate-level knowledge-based test. On other benchmarks, Claude 3's improvements over GPT-4 are more impressive, including several percentage-point improvements in graduate-level reasoning, mathematical reasoning, and coding proficiency, as shown below.
Source: Anthropic 2024.1 For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results.
Anthropic is tiering its Claude 3 API (application programming interface) prices, as shown below. The price of Claude 3 Opus, its highest performing API, is roughly twice that of GPT-4, at a weighted average of $30 per million tokens,2 while that of Sonnet, its mid-tier version, is less than GPT-4 at $6 per million tokens. For perspective, GPT-4 Turbo costs $15 per million tokens on a weighted average basis.3
Source: ARK Investment Management LLC, 2024, Anthropic 2024,4 OpenAI 2024. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results.
Claude 3’s broad-based capabilities are a reminder that model performance should be measured against a wide range of benchmarks, and their utility evaluated on the basis of real-world use cases.
2. Rivian Announced R2 Delivery Plans But Paused Construction Of Its Electric Truck Plant
On Thursday, Rivian announced that it will deliver its next vehicle, the R2, in the first half of 2026. The R2 will cost $45,000 and have ~300 miles of range.5 Rivian also announced that it is pausing construction of a $5 billion plant in Georgia, which will cut its annual capital spending budget by ~$2.25 billion... "for now."6
These announcements highlight Rivian’s quandary: customers seem to love the R2 depicted below, but production costs still are so high that profitability remains elusive. We would not be surprised to see Rivian try and raise additional funds in the near term as it pivots in an attempt to ramp production of a profitable mass market vehicle during the next few years.
Source: Rivian 2024.7 For illustrative purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.
3. Starlink's Direct-To-Device Service Is On Track For US Launch Later This Year
Last week, SpaceX reported that it is on track to commercialize a direct-to-device Starlink service in the US later this year.8 Since initial testing in January, the company has sent and received traditional SMS and WhatsApp text messages and posted on X using unmodified devices. Interestingly, the devices have worked indoors, perhaps because Starlink has been testing at a lower earth orbit.
According to our research, when SpaceX and T-Mobile launch direct-to-device service, satellite subscribers as a percent of cellular subscribers could increase from ~0.04% today to ~1.25% during the next four to five years, as shown below.9 As more cellular carriers incorporate direct-to-device satellite service, emergency SOS could become universal long term, supported by Starlink, Iridium,10 AST SpaceMobile,11 as well as the new Mobile Satellite Services Association, including Viasat,12 and others.
Source: ARK Investment Management LLC, 2024. This ARK forecast is based on a range of data sources, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results.
1 Anthropic 2024. “Introducing the Next Generation of Claude.”
2 Assuming three input tokens are fed into the model for every one output token produced.
3 OpenAI. 2024. “Pricing.” Importantly, some note that Anthropic has produced a model generally more performant on a benchmark basis than Google, which has access to more resources including data, compute, and talent. See Levit, C. 2024. “It is flabbergasting that google…” X.
4 ARK Investment Management LLC, 2024. “Big Ideas 2024: Disrupting The Norm, Defining The Future.” Anthropic. 2024. “Pricing.” OpenAI. 2024. “Pricing.”
5 Rivian. 2024. “Rivian R2.”
6 Amy, J. and Bynum, R. 2024. “Automaker Rivian pauses construction of its $5 billion electric truck plant in Georgia.” The Associated Press.
7 Rivian. 2024. “Rivian R2.”
8 SpaceX. 2024. “Dear Ms. Dortch.” Letter to the Federal Communications Commission, March 7, 2024.
9 ARK Investment Management LLC, 2024. “Big Ideas 2024: Disrupting The Norm, Defining The Future.”
10 Iridium. 2024. “Iridium Unveils Project Stardust.”
11 AST Space Mobile. 2024. “SpaceMobile Network Direct Connection.”
12 Rabinow, J. 2024. “Satellite operators join forces to chase direct-to-smartphone opportunity.” Space News.