Mira Murati’s Inkling AI Model Review: Best Open-Source Model
After two years of silence from Thinking Machines Lab, Murati's debut model is out and on OpenRouter. The MCP score is genuinely impressive. The price-to-performance math is more complicated.
TL;DR
OpenAI and Anthropic dramatically reduced prices on mid‑tier AI models in 2026, aiming to outpace open‑source and Chinese competitors. The cuts lower costs for developers but compress margins as both firms prepare for IPOs.
Key facts
- OpenAI cut GPT‑5.6 Luna price by 80% to $0.20 per million input tokens on July 30.
- OpenAI reduced GPT‑5.6 Terra price by 20% to $2 per million input tokens.
- Anthropic set permanent pricing for Claude Sonnet 5 at $2/$10 per million tokens on August 10.
- OpenAI launched GPT‑6 Luna at $0.10 per million input tokens and $0.50 per million output tokens on September 22.
- Anthropic introduced Claude Opus 5.5 at $4/$20 per million tokens, promising about 40% effective savings.
OpenAI and Anthropic have launched a series of steep price reductions on their mid‑tier AI models throughout 2026, aiming to outpace open‑source alternatives and Chinese competitors. OpenAI cut its GPT‑5.6 Luna model by 80% on July 30, setting the price at $0.20 per million input tokens and $1.20 per million output tokens, while also lowering GPT‑5.6 Terra by 20%. Anthropic followed with permanent pricing for Claude Sonnet 5 on August 10 at $2/$10 per million tokens and later introduced Claude Opus 5.5 at $4/$20 list pricing, promising effective savings of about 40%. The moves come as both firms prepare for potential IPOs and confront aggressive pricing from Chinese AI firms such as DeepSeek.
Background
The AI market in 2026 has become a battleground of pricing strategies, with developers weighing the cost of proprietary APIs against the operational overhead of self‑hosting open‑source models. Historically, open‑source offerings have attracted a sizable share of the market, especially on aggregators like OpenRouter where users can mix and match models. However, the arrival of lower‑priced proprietary options is shifting that balance. Chinese AI companies have entered the fray by offering models such as DeepSeek V4 Flash at roughly $0.14 per million input tokens and $0.28 per million output tokens, undercutting U.S. pricing and narrowing the performance gap. Both OpenAI and Anthropic are widely expected to pursue IPOs, and their aggressive pricing is intended to boost adoption metrics and usage volume, even as it compresses per‑call margins. The broader context includes enterprises segmenting workloads: cheaper models handle routine tasks, while premium models are reserved for specialized, high‑performance needs. This tiered approach influences how price cuts affect overall market dynamics and developer choices.
What happened
On July 30, OpenAI announced an 80% reduction in the price of its GPT‑5.6 Luna model, bringing the cost down to $0.20 per million input tokens and $1.20 per million output tokens, as reported by CryptoBriefing. In the same announcement, the company also reduced the price of its higher‑end GPT‑5.6 Terra model by 20%, setting it at $2 per million input tokens and $12 per million output tokens. Anthropic responded on August 10 by making the introductory pricing for Claude Sonnet 5 permanent at $2 per million input tokens and $10 per million output tokens. Later, on September 22, OpenAI launched two new models: GPT‑6 Sol at $2 per million input tokens and $10 per million output tokens, and GPT‑6 Luna at $0.10 per million input tokens and $0.50 per million output tokens, representing roughly a 50% reduction compared with the previous generation. Anthropic countered with Claude Opus 5.5, listed at $4 per million input tokens and $20 per million output tokens, while noting that built‑in efficiency improvements deliver around 40% effective savings for users. Throughout these announcements, CryptoBriefing highlighted that Chinese competitors such as DeepSeek V4 Flash are pricing their models at roughly $0.14 per million input tokens and $0.28 per million output tokens, further intensifying the pricing pressure on U.S. providers.
Market impact
The price cuts have reshaped cost calculations for developers. OpenAI's GPT‑5.6 Luna now costs $0.20 per million input tokens, a dramatic drop from its prior level, and GPT‑5.6 Terra sits at $2 per million input tokens after a 20% reduction. Anthropic's Claude Sonnet 5 permanent pricing of $2/$10 per million tokens offers a stable baseline for users. The September launch of GPT‑6 Luna at $0.10 per million input tokens and $0.50 per million output tokens makes it the cheapest U.S. offering, while GPT‑6 Sol at $2/$10 mirrors the pricing of Claude Sonnet 5. Anthropic's Claude Opus 5.5, listed at $4/$20, claims effective savings of about 40% due to efficiency gains. Chinese model DeepSeek V4 Flash, priced at $0.14/$0.28, remains cheaper than many U.S. options, though the performance gap is narrowing. These shifts are expected to drive higher API call volumes, potentially offsetting thinner margins as both OpenAI and Anthropic prepare for IPOs.
Why it matters
The aggressive pricing strategy signals that leading U.S. AI firms are willing to sacrifice short‑term margin to capture market share from open‑source and Chinese providers. By lowering the cost barrier, they aim to increase the volume of API calls, which could improve revenue stability ahead of anticipated IPOs. The price competition also forces developers to reassess the trade‑off between self‑hosting open‑source models and using proprietary APIs, potentially accelerating the adoption of cloud‑based AI services. Moreover, the narrowing performance gap with Chinese models suggests a more globally competitive landscape, where cost and efficiency become primary differentiators. These dynamics may reshape funding flows, talent allocation, and the strategic focus of AI startups that rely on external model access.
DeFi Intel’s take
Editorial analysis, not a statement of fact.
DeFi Intel's take: the price war initiated by OpenAI and Anthropic suggests a strategic pivot toward volume‑driven growth as both firms eye public listings. We read this as an effort to lock in developer loyalty before open‑source and Chinese alternatives become more capable and cost‑effective. The sustained low‑price tiers could compress margins for the incumbents, but the anticipated surge in usage may offset that pressure. If Chinese models continue to close the performance gap, the competitive advantage of U.S. providers may shift from raw capability to ecosystem integration and reliability. This environment could also influence crypto‑related AI projects, which often balance cost against decentralization, prompting a re‑evaluation of model sourcing decisions.
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 price did OpenAI set for GPT‑5.6 Luna after the July 30 cut?
OpenAI reduced the price of GPT‑5.6 Luna to $0.20 per million input tokens and $1.20 per million output tokens, representing an 80% cut from its previous pricing.
How does Anthropic's Claude Opus 5.5 pricing compare to its listed price after efficiency gains?
Claude Opus 5.5 is listed at $4 per million input tokens and $20 per million output tokens, but Anthropic notes that built‑in efficiency improvements deliver around 40% effective savings for users.
Why are Chinese AI models like DeepSeek V4 Flash relevant to the pricing war?
Chinese models such as DeepSeek V4 Flash are priced at roughly $0.14 per million input tokens and $0.28 per million output tokens, undercutting U.S. offerings and narrowing the performance gap, which adds pressure on OpenAI and Anthropic to lower their prices.
Direct answers
Short answers, written to be quoted.
- What are the new prices for OpenAI's GPT‑5.6 Luna model?
- The new price for GPT‑5.6 Luna is $0.20 per million input tokens and $1.20 per million output tokens after an 80% reduction.
- Which Chinese AI model is undercutting U.S. pricing and at what cost?
- DeepSeek V4 Flash offers pricing of roughly $0.14 per million input tokens and $0.28 per million output tokens, undercutting U.S. models.
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 Decrypt.
Discussion
Comments are reviewed before publication.
Want the full article?
Continue reading on Decrypt →
No comments yet.