OpenAI cut developer prices for its flagship GPT-5.6 Sol model by more than 20% on Friday, dropping output tokens to $20 per million for three months. Key Points: GPT-5.6 Sol now runs $4 per
OpenAI cut developer prices for its flagship GPT-5.6 Sol model by more than 20% on Friday, dropping output tokens to $20 per million for three months.
- GPT-5.6 Sol now runs $4 per million input tokens and $20 per million output tokens, down from $5 and $30.
- The lower rates cover the API plus credits for ChatGPT Work and Codex, while Pro, Plus and Business subscriptions stay put.
- Promotional pricing holds at least through Nov. 21, landing as average enterprise inference costs sink to a 2026 low.
The company detailed the change on Aug. 21, applying the reduced rates to standard short-context use on the application programming interface where it publishes its rate card.
Cached input slid to $0.40 per million tokens from $0.50. That preserves the tenfold discount developers earn for reusing the same context across repeated calls.
Credits for ChatGPT Work, the agentic product, and Codex, the coding tool, both qualify for the lower rates, while Pro, Plus and Business subscribers pay the same as before. Sol remains the priciest model in the OpenAI lineup, and its promotional rate runs at least through Nov. 21, against sharpening competition from Anthropic and Chinese labs.
Also Read:Sandbox Contains Exploit After Attacker Mints 14.9B SAND Tokens
Investment bank Jefferiesmeasured average inference prices between $1.16 and $1.18 per million tokens from Aug. 6 to Aug. 8, the lowest reading recorded this year on an index tracking commercial providers and open-weight platforms. Analysts led by Thomas Chong tied the decline to tighter cost discipline across U.S. and Chinese technology firms, alongside surging adoption of cheap open-weight models such as DeepSeek. Unit costs had stood at $2.04 on May 31 and $1.45 in late July.
Sol's new $4 input rate now undercuts Anthropic's Claude Opus 5, which charges $5 per million input tokens and $25 per million output, leaving the discounted OpenAI model cheaper on both sides of the ledger. Its costlier Claude Fable 5 lists at $10 and $50 per million tokens.
Buyers increasingly shop on price because the performance gap between leading models keeps narrowing, a pattern that former OpenAI go-to-market chief Zack Kasscalls diminishing model returns, and one that erodes any single lab's power to command a premium.
OpenAI reduced rates across the rest of the GPT-5.6 line on Jul. 30, roughly three weeks after the Terra and Luna models reached the public. Terra fell 20% to $2 per million input tokens. Luna dropped 80% to $0.20, the steepest cut in the family.
The company credited that round to production GPU kernel improvements and better speculative decoding, gains it said came from tasking GPT-5.6 with optimizing its own runtime. Sol sat out that round entirely, and chief executive Sam Altman has argued that sheer usage volume, rather than fat margins, will pay for training each new generation of frontier models.
Read Next:Bitcoin Slides From $79.5K As RSI Hits Seven-Year Extreme
