A price war has broken out among the leading artificial intelligence labs, with Anthropic and OpenAI cutting their prices within 36 hours of each other this week — while xAI moved in the opposite direction and raised prices for long prompts.
The cuts underline a broader trend: writing code with AI keeps getting cheaper, and the labs are competing aggressively for developer workloads.
Anthropic cuts cache reads by 60%
Anthropic reduced the price of cache reads on its Claude Opus 5.5 model by 60 per cent, from $0.50 to $0.20 per million tokens. Cache reads are what AI agents pay to reuse context they have already processed, so long coding sessions over the same repository collect the discount on every turn.
The company also cut input pricing from $5 to $4 per million tokens and output from $25 to $20, and says output speed is more than 30 per cent faster than the previous Opus 5. Anthropic reported that one tester completed a 680,000-line code migration in under a day, work the company says would have taken a team weeks.
Claude Code also made Opus the default model on Pro and Team Standard plans, replacing Sonnet, and raised usage limits on paid plans.
OpenAI halves GPT-6 series pricing
OpenAI responded on September 22 with GPT-6 Sol and GPT-6 Luna at half the cost of the 5.6 series, savings the company credits to improvements in caching and inference. Sol is positioned as the coding and agent model, while Luna handles high-volume clerical work such as summarising documents and extracting fields.
xAI raises Grok prices for long prompts
Elon Musk's xAI went the other way. Grok 4.7 now charges double for prompts above 200,000 tokens — $4 per million input tokens versus $2 below that threshold — while its 500,000-token context window is unchanged. That makes long-context agent work the one AI task that got more expensive this week.
Tools move beyond writing code
The price cuts arrived alongside a shift in developer tooling toward reviewing and deploying rather than just writing code. Cursor released two bots: Rollouts, which follows a change from pull request through deployment and compares post-deploy metrics against the baseline to catch regressions early, and Security Review, which scans pull requests for injection flaws and committed secrets with one-click fixes.
The company reported that AI-assisted review cut average review time from 4.8 to 3.8 minutes, with reviewers accepting a higher share of the bots' comments.
Sources: EveryDev.ai — Weekly AI Dev News Digest: September 19–25, 2026 — https://www.everydev.ai/p/news-weekly-ai-dev-news-digest-september-19-25-2026
The cuts underline a broader trend: writing code with AI keeps getting cheaper, and the labs are competing aggressively for developer workloads.
Anthropic cuts cache reads by 60%
Anthropic reduced the price of cache reads on its Claude Opus 5.5 model by 60 per cent, from $0.50 to $0.20 per million tokens. Cache reads are what AI agents pay to reuse context they have already processed, so long coding sessions over the same repository collect the discount on every turn.
The company also cut input pricing from $5 to $4 per million tokens and output from $25 to $20, and says output speed is more than 30 per cent faster than the previous Opus 5. Anthropic reported that one tester completed a 680,000-line code migration in under a day, work the company says would have taken a team weeks.
Claude Code also made Opus the default model on Pro and Team Standard plans, replacing Sonnet, and raised usage limits on paid plans.
OpenAI halves GPT-6 series pricing
OpenAI responded on September 22 with GPT-6 Sol and GPT-6 Luna at half the cost of the 5.6 series, savings the company credits to improvements in caching and inference. Sol is positioned as the coding and agent model, while Luna handles high-volume clerical work such as summarising documents and extracting fields.
xAI raises Grok prices for long prompts
Elon Musk's xAI went the other way. Grok 4.7 now charges double for prompts above 200,000 tokens — $4 per million input tokens versus $2 below that threshold — while its 500,000-token context window is unchanged. That makes long-context agent work the one AI task that got more expensive this week.
Tools move beyond writing code
The price cuts arrived alongside a shift in developer tooling toward reviewing and deploying rather than just writing code. Cursor released two bots: Rollouts, which follows a change from pull request through deployment and compares post-deploy metrics against the baseline to catch regressions early, and Security Review, which scans pull requests for injection flaws and committed secrets with one-click fixes.
The company reported that AI-assisted review cut average review time from 4.8 to 3.8 minutes, with reviewers accepting a higher share of the bots' comments.
Sources: EveryDev.ai — Weekly AI Dev News Digest: September 19–25, 2026 — https://www.everydev.ai/p/news-weekly-ai-dev-news-digest-september-19-25-2026
