OpenAI and Anthropic Slash Token Prices Amid Competition

Anthropic launched Claude Sonnet 5 on June 30, 2026, pricing it at an introductory $2 per million input tokens and $10 per million output tokens through August 31 (standard pricing of $3/$15 after that), compared with Opus 4.8's $5 input/$25 output rate; Anthropic says the introductory pricing is set to make the transition roughly cost-neutral despite a new tokenizer that uses more tokens per input. Separately, the Wall Street Journal reported on June 11, citing people familiar with the matter, that OpenAI is weighing steep cuts to its own token prices and expects Anthropic to follow suit; CNBC and Bloomberg picked up the report the same day, noting it surfaced shortly after both companies had filed confidential IPO paperwork. As of this writing, OpenAI has not announced specific price changes. Reuters reported July 2 that an inexpensive new Chinese AI model is gaining ground on both companies in their core markets, adding a third source of pricing pressure. Separately, the-decoder reported Anthropic pulled back a planned billing overhaul for the Claude Agent SDK as the price competition intensified.

1. June 11, 2026

   WSJ reports OpenAI is weighing steep token-price cuts in anticipation of Anthropic doing the same.

2. June 30, 2026

   Anthropic launches Claude Sonnet 5 at introductory pricing of $2/$10 per million input/output tokens.

3. July 2, 2026

   Reuters reports a new, inexpensive Chinese AI model is gaining ground on both companies.

Sonnet 5's confirmed pricing is the more actionable number today: at introductory rates it is meaningfully cheaper per token than Opus 4.8, and Anthropic says the tokenizer change keeps the transition close to cost-neutral, so teams already on Sonnet 4.6 should re-benchmark accuracy and latency against Sonnet 5 rather than assuming price alone justifies switching. If OpenAI's reported cuts do materialize, expect the more consequential engineering shift to be architectural rather than just cheaper API calls: lower per-token costs raise the relative importance of request batching, prompt caching, and context-length management, since those levers now determine a larger share of total inference spend once the sticker price drops.

Watch for an official OpenAI pricing announcement, none has been made as of July 2, and for whether it undercuts Sonnet 5's $2/$10 introductory rate or targets a different tier. Also track how the reported Chinese low-cost entrant performs on real workloads rather than benchmarks, and whether Anthropic or OpenAI adjust IPO-related disclosures given how directly pricing strategy affects near-term revenue guidance.