Anthropic reveals internal reasoning steps of Claude model, and
Anthropic revealed internal reasoning steps of Claude that resemble human brain structure, potentially enhancing AI transparency and safety.
TL;DR
In September 2026 OpenAI, Anthropic and xAI launched cheaper frontier AI models, cutting token prices by 40‑50 % to counter Chinese competitors and create space for safety work. The move reshapes AI economics, pressures mid‑tier firms and forces cloud providers to adjust margins.
Key facts
- OpenAI’s GPT‑6 Luna tier costs $0.10 per million input tokens and $0.50 per million output tokens.
- Anthropic’s Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens.
- xAI’s Grok 4.7 launched at $2 per million input tokens and $6 per million output tokens.
- Chinese AI models were undercutting U.S. offerings by up to nine times on a per‑token basis.
- Sam Altman, Dario Amodei and Elon Musk called for a slowdown in frontier AI development in 2026.
In September 2026 OpenAI, Anthropic and Elon Musk’s xAI each launched new, cheaper frontier AI models, marking a sharp drop in per‑token pricing. OpenAI introduced GPT‑6’s Sol and Luna tiers, Anthropic released Claude Opus 5.5 and xAI unveiled Grok 4.7. The moves came as Chinese competitors were offering token prices up to nine times lower, prompting the three U.S. labs to cut costs by roughly 40‑50 percent to keep pace and give themselves breathing room for safety work.
Background
The AI market in mid‑2026 was under intense pressure from overseas rivals, especially Chinese firms that were pricing comparable large‑language models at a fraction of U.S. rates. Industry leaders such as Sam Altman of OpenAI, Dario Amodei of Anthropic and Elon Musk of xAI publicly urged a slower pace of frontier AI development so safety research and regulation could catch up. Amodei warned that autonomous AI agents could become a near‑term risk within six to twelve months if safeguards lagged. At the same time, the industry’s shift to usage‑based pricing meant that revenue now follows the volume of tokens processed rather than fixed licences, tying lab earnings directly to how much developers query the models. Cloud providers like AWS, Google Cloud and Azure, which host managed inference services, also felt the ripple effect because lower model costs reshape the margin structure of the entire inference stack. The convergence of competitive pricing pressure, safety concerns and a usage‑centric business model set the stage for the September price cuts.
What happened
The September 2026 releases are detailed in a Cryptobriefing report. OpenAI’s GPT‑6 lineup debuted two tiers: Sol, priced at about $2 per million input tokens and $10 per million output tokens, and Luna, priced at $0.10 per million input and $0.50 per million output tokens, representing roughly a 50 % cut from the prior GPT‑5.6 pricing. Anthropic’s Claude Opus 5.5 arrived with a price of $4 per million input tokens and $20 per million output tokens, a reduction of about 40 % from its predecessor. xAI’s Grok 4.7 launched around September 21‑22, 2026, with pricing set at $2 per million input tokens and $6 per million output tokens, emphasizing token efficiency as its main selling point. The report notes that Chinese AI models were undercutting U.S. offerings by as much as nine times on a per‑token basis, a factor that forced the price adjustments. The three CEOs—Altman, Amodei and Musk—each made public statements in 2026 calling for a deliberate slowdown in frontier AI development to allow safety research and regulatory frameworks to catch up. Amodei specifically highlighted autonomous AI agents as a near‑term risk, warning that dangerous capability gaps could emerge within a six‑to‑twelve‑month window if safeguards were not in place.
Market impact
The new pricing levels translate into significant cost savings for developers. OpenAI’s Luna tier at $0.10 per million input tokens and $0.50 per million output tokens is dramatically cheaper than its own Sol tier and far below the $2‑$10 rates of the previous generation. Anthropic’s Claude Opus 5.5 at $4 input and $20 output per million tokens cuts its own costs by roughly 40 %. xAI’s Grok 4.7 at $2 input and $6 output per million tokens also undercuts its earlier models. These reductions compress the price gap between frontier and commodity models, pressuring mid‑tier AI firms that relied on that spread to justify higher fees. Cloud providers that bundle inference services must renegotiate margins, as lower model costs ripple through the entire stack. The competitive pressure from Chinese models, which were priced up to nine times lower, is a key driver behind the U.S. labs’ decision to slash prices, aiming to retain enterprise customers who might otherwise migrate to cheaper alternatives.
Why it matters
The price cuts matter because they reshape the economics of AI adoption and safety research. By lowering per‑token costs, OpenAI, Anthropic and xAI give themselves financial breathing room to slow development and focus on safety, aligning with the public calls from Altman, Amodei and Musk for a more deliberate pace. The reduced cost also makes near‑frontier models accessible to a broader range of developers, potentially accelerating the diffusion of powerful AI capabilities across more applications. At the same time, the narrowing of the price gap between top‑tier and mid‑tier models forces smaller AI firms to justify higher fees, potentially leading to consolidation or a shift toward niche specialisation. Cloud providers will need to adjust pricing and margin structures, which could affect the overall profitability of AI‑as‑a‑service offerings. Finally, the competitive pressure from Chinese models underscores a global pricing race that could influence where AI research and deployment concentrate in the coming years.
DeFi Intel’s take
Editorial analysis, not a statement of fact.
DeFi Intel's take: the September price cuts suggest a strategic pivot by the three leading labs to balance market share with safety imperatives. By making frontier models cheaper, they appear to be buying time for safety research while staving off a migration to far cheaper Chinese alternatives. We read this as a signal that usage‑based pricing will continue to dominate, forcing cloud providers and mid‑tier AI firms to adapt quickly. The narrowing price gap may also accelerate the integration of advanced AI into DeFi protocols, as lower costs lower the barrier for on‑chain AI services. However, the pressure on margins could spur further consolidation in the AI stack, potentially limiting the diversity of models available to developers.
Sources used for this analysis
DeFi Intel wrote this analysis from the reporting above. Any figure or quotation can be checked against those pages.
Frequently asked questions
What new AI models were launched in September 2026?
OpenAI launched GPT‑6 with Sol and Luna tiers, Anthropic released Claude Opus 5.5, and xAI introduced Grok 4.7. Each model was priced lower than its predecessor, aiming to make near‑frontier performance more affordable for developers.
How do the new token prices compare to the previous generation?
OpenAI’s Luna tier costs $0.10 per million input tokens and $0.50 per million output tokens, about a 50 % cut from GPT‑5.6. Anthropic’s Claude Opus 5.5 is priced $4 input and $20 output per million tokens, roughly a 40 % reduction. xAI’s Grok 4.7 is set at $2 input and $6 output per million tokens, also lower than earlier rates.
Why did the labs decide to lower their prices?
The price reductions were driven by intense competition from Chinese AI models that were up to nine times cheaper, as well as public calls from industry leaders to slow frontier AI development for safety reasons. Lower costs give labs financial breathing room to focus on safety while retaining market share.
Direct answers
Short answers, written to be quoted.
- What are the new token prices for OpenAI’s GPT‑6 models?
- OpenAI’s GPT‑6 Luna tier is priced at $0.10 per million input tokens and $0.50 per million output tokens, while the Sol tier costs about $2 per million input tokens and $10 per million output tokens.
- How much cheaper is Anthropic’s Claude Opus 5.5 compared to its predecessor?
- Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens, representing roughly a 40 % reduction from the previous model’s rates.
- Why did the AI labs lower their model costs in September 2026?
- The labs reduced prices to counter Chinese AI models that were up to nine times cheaper and to create financial breathing room for safety research after public calls from Altman, Amodei and Musk to slow frontier AI development.
DeFi Intel is an entity-graph aggregator: we curate, tag and link crypto news to a typed knowledge graph of protocols, tokens, people and incidents. We do not republish the full article body. Use the link above to read the original report at Crypto Briefing.
Anthropic's research on Claude's reasoning structure was reported by Crypto Briefing, linking the AI lab's findings on model transparency to the publication's coverage.
Discussion
Comments are reviewed before publication.
Want the full article?
Continue reading on Crypto Briefing →
No comments yet.