OpenAI and Anthropic Cut LLM Prices Up to 80%

For coding agents, cache reads make up 96% to 99% of total token spending, because the cache holds the entire previous conversation on every turn. The cache is hit roughly 90% to 100% of the time depending on the inference provider, so a 75% cut in cache read cost translates into a 70-75% cut in the cost of using a model.

That is why Anthropic's Fable 5.1, with cache reads at $0.25, matters more than its headline input and output price suggests. OpenAI matched the same 5%-of-input cache read ratio with GPT-6.1 Sol at $0.10, making that model family cumulatively 75% cheaper, per the source.

No one cuts prices by 80% without an outside reason. The source argues that models have become a commodity: developers can swap one for another with a toggle in tools like OpenCode and Zed Agent, or by changing an API endpoint string through OpenRouter. That leaves little moat for any single provider.

Competition adds pressure. Open-weight models such as DeepSeek Flash V4.1, whose updated 1017 variant recently shipped, offer sufficient substitutes, and larger players like Google and Meta can supply comparable models. The source notes that without its 80% Luna cut, OpenAI would have lost to DeepSeek V4 Pro on cost-per-task. Data from Vercel's AI model gateway shows cost-per-token falling to 43.3% of the April 2026 level, even as total spend rises because models improve faster than they get cheaper.

For developers and startups in Tunisia, the practical effect is that building with frontier models costs far less than it did in June. Small teams can run agentic coding workflows on GPT-6.1 Sol or Claude Opus 5.5 without a large budget, and the same cuts reach subscription tools like Codex and Claude Code.

The source reports no drop in output quality to achieve these prices, though it flags mixed real-world feedback on some releases. Pricing was published in US dollars; at roughly 2.9 TND per dollar, GPT-6.1 Sol's $2 input price is about 5.8 TND per million tokens. No Tunisian availability or local pricing has been announced, so teams should check current rates directly with each provider.