OpenAI and Anthropic Execute Successive 20-80% LLM Price Cuts Across GPT-5.6 to GPT-6.1 and Claude Opus 5.5 Families June-September 2026

OpenAI announced GPT-5.6 Sol and Luna on June 26 at $5/$30/$0.50 and $1/$6/$0.10. Anthropic followed with Claude Sonnet 5 at $2/$10/$0.20 on June 30 and Opus 5 at $5/$25/$0.50 on July 24. OpenAI then cut Luna 80% on July 30, made Sonnet 5 cuts permanent August 10, reduced Sol 20% August 21, released GPT-6 Astra September 3, and dropped GPT-6 Sol to $2/$10/$0.20 plus GPT-6.1 Sol cache to $0.10 on September 22-29. Anthropic released Fable 5.1 September 1 with 75% lower cache reads and Opus 5.5 at $4/$20/$0.20 September 22.

Cache reads comprise 96-99% of agent token spend with 90-100% hit rates, so the 75% cache cut in Fable 5.1 and GPT-6.1 Sol produces 70-75% total cost drops. Benchmark gains in Opus 5.5 and GPT-6.1 Sol occurred alongside explicit compute-efficiency statements from both labs. These moves track inference optimization patterns seen in prior generations rather than pure model scaling.

Price compression directly lowers barriers for agentic coding and multi-turn workflows that previously exceeded $10 per task. Sustained 40-75% reductions across input, output, and cache metrics accelerate experimentation in regulated sectors where per-token costs previously constrained deployment. Operational impact appears in expanded context windows and higher cache reuse rates without proportional spend increases.

Next quarter will test whether these cuts stabilize or accelerate further as GPT-6.1 and Opus 5.5 inference footprints propagate to additional providers. Sustained sub-$2 input pricing for mid-tier models is likely to shift evaluation focus from raw capability to cost-per-successful-task metrics.