OpenAI and Anthropic Cut Artificial Intelligence Model Prices by Up to 80 Percent

Over the past four months, OpenAI and Anthropic lowered the prices of their flagship artificial intelligence models by as much as 80%. This shift occurs as growing pressure from affordable open-weight alternatives and upcoming public market discussions challenge both organizations.

Starting in June, these fast-paced reductions dropped API expenses by 75% or higher in certain situations. Rival firms like DeepSeek and Moonshot AI put additional pressure on the two market leaders by distributing inexpensive models via platforms such as OpenRouter, where numerous suppliers host artificial intelligence inference services.

Major Price Cuts

OpenAI launched this wave of cuts in late June by introducing GPT-5.6 Luna at $1 per million input tokens and $6 per million output tokens. By July 30, the firm slashed both rates by 80% down to $0.20 and $1.20.

Anthropic responded in August by turning its temporary introductory discount for Claude Sonnet 5 into a permanent rate. In September, the business further trimmed pricing for Claude Opus 5.5 by 40% to 60%.

Days afterward, OpenAI revealed GPT-6.1 Sol with input and output costs set at half of its predecessor. The enterprise also dropped cached input rates to 5% of normal input expenses, matching Anthropic.

Max Woolf, a former senior data scientist at BuzzFeed who monitors artificial intelligence pricing on his personal blog, expressed astonishment at how quickly these drops occurred. He mentioned that leading model expenses have fallen much faster than anticipated a year prior.

Reduced rates matter significantly for programming helpers like Codex and Claude Code. Such utilities repeatedly query stored conversation logs, where cache reads make up 96% to 99% of token expenditure during tasks.

Consequently, a 75% drop in cache-read fees diminishes active usage costs by roughly 70% to 75%.

Metrics from Vercel's artificial intelligence gateway indicate that the mean expense per token dropped to 43.3% of April 2026 levels. Even so, overall expenditure keeps climbing because better model abilities prompt heavier utilization.

Cheaper Open-Weight Models Increase Competition

OpenAI and Anthropic also confront heavier competition from open-weight systems, specifically those built by DeepSeek and Moonshot AI.

DeepSeek introduced its Flash V4.1 1017 model in April, setting input, output, and cached token costs at $0.15, $0.60, and $0.003 per million tokens.

Buyers can now access this system through 29 distinct inference suppliers on OpenRouter, up from 13 in May. Vendors like Cloudflare and DeepInfra deliver active usage expenses within 10% of DeepSeek's native endpoint.

Concurrently, Moonshot AI's Kimi K3.1 Pro reportedly matches Claude Opus 5.5 performance while costing half the input and output price. Moonshot AI schedules the release of model weights for November 4.

Cost evaluations demonstrate that OpenAI's 80% discount for GPT-5.6 Luna greatly enhanced its financial standing against DeepSeek V4 Pro. Absent the reduction, DeepSeek's option would have secured a superior position on quality-versus-expense comparisons.

Despite reduced rates and wider availability on OpenRouter, open-weight models still represent a small share of total token consumption.

Woolf stated that OpenAI and Anthropic must keep building superior models to rival inexpensive choices. He pointed to Claude Opus 5.5 as an illustration of how premium systems retain an edge through performance.

Price Cuts Raise Questions About Potential IPOs

These reductions coincide with reports that both businesses weigh entries into the public market.

Rumors suggest OpenAI eyes an initial public offering in 2027, whereas Anthropic might file for an IPO within several months.

Both enterprises disclosed billions of dollars in losses linked to research, development, and infrastructure scaling. Yet those figures predate the recent surge in artificial intelligence agent adoption.

Trimming prices by up to 80% might complicate financial standings before potential public offerings, especially if lower rates eat into profit margins.

Nevertheless, certain reports indicate the firms earlier maintained profit margins of approximately 80% per token prior to recent drops, potentially providing room to lower prices while staying profitable on inference.

These adjustments also altered how programmers handle artificial intelligence budgets.

Woolf shared that GPT-6.1 Sol pricing dropped his active usage expenses by roughly 75%. He subsequently reduced his Codex usage limit to one-fifth of prior levels while maintaining steady total expenditures.

Simultaneously, OpenAI reduced usage allowances for its $200-per-month Codex tier by 50%, pointing to increased model efficiency.

AI Model Prices Could Fall Further

Woolf predicts that artificial intelligence pricing may keep dropping as long as rival inference providers sustain matching performance and rates while enabling simple customer migration.

Meanwhile, OpenAI and Anthropic attempt to strengthen client bonds via corporate contracts and personal assistants, including OpenAI Dots and Claude Duet.

Still, climbing rivalry could eventually push rates closer to the exact expenses required to operate artificial intelligence systems.

Should current patterns persist, programmers could enjoy additional decreases in utilization costs, while major artificial intelligence entities face mounting pressure to boost performance and preserve profitability.