OpenAI and Anthropic now price their mainline models identically. GPT-6.1 Sol and Claude Sonnet 5.5 both list at $2 per million input tokens and $10 per million output, and the budget models, GPT-6 Luna and Claude Haiku 5.5, both sit at $0.10 and $0.50. Neither gives a new account a free credit. So the per-token table settles nothing, and the decision comes down to caching, rate limit tiers and account policy. Prices checked on both companies’ own pages on November 2, 2026.
The real differences show up once you look past the headline rates. Anthropic charges a premium to write a cache but discounts reads more deeply on its newest models. OpenAI opens a $200,000 monthly cap after $500 of purchases, while Anthropic sets the same cap on its Scale tier but starts new organizations in a throttled Evaluation tier. Those two details, not the list price, decide the bill for a production app.
Don’t expect a free credit from either
Neither company promises a signup credit. OpenAI has a Free usage tier in its rate limit docs, but that is a $100 monthly spending ceiling for organizations in an allowed geography, not a grant, and moving to the Build tier requires $5 in total credit purchases. Anthropic sells prepaid credits that expire one year after purchase and says nothing about a free balance for new Console organizations. Budget a small purchase on either and move on.
Where Anthropic does hand out API usage is through its consumer and team subscriptions. A Max plan includes $100 per month in API credits, Ultra includes $200, and Team includes up to $500 per month pooled across the organization. Those monthly credits are spent before purchased credits, expire at the end of each billing cycle, and can’t be used in Claude Code or the Claude apps. If your team already pays for Max, that’s $1,200 a year of API usage you’d otherwise be buying.
Where the two bills start to diverge
The per-token tables match almost line for line. GPT-6.1 Sol and Claude Sonnet 5.5 are $2 in and $10 out, GPT-6 Luna and Claude Haiku 5.5 are $0.10 and $0.50, and the two top models, GPT-6 Astra and Claude Fable 5.1, are $10 and $50. One wrinkle on the Anthropic side: Haiku 5.5 jumps to $0.50 and $2.50 once a prompt passes 100,000 tokens, five times the base rate, so long-context work on the budget model isn’t cheap.
Caching is the fork in the road. OpenAI charges cached input at $0.10 per million on Sol and $0.01 on Luna with no write fee. Anthropic charges 1.25 times the input price to write a 5 minute cache and 2 times for a 1 hour cache, then reads at 0.05 times input on Sonnet 5.5 and Opus 5.5 and 0.025 times on Fable 5.1. A prompt read many times costs less on Anthropic. A prompt read once costs more.
Both offer a 50% batch discount. OpenAI’s Batch API completes within 24 hours and draws from a separate queue limit, and its Flex processing bills at batch rates for slower, interruptible requests. Anthropic’s Message Batches API has its own rate limits, 1,000 requests per minute and 200,000 queued requests on the Start tier, which is plenty for a nightly job.
How fast your limits grow
OpenAI collapsed its paid tiers to three in 2026. Build opens at $5 in cumulative purchases with a $500 monthly limit, Launch at $100 with $5,000, and Grow at $500 with $200,000. On Build, the Sol and Astra family gets 5,000 requests and 1 million tokens per minute, and Luna gets 2 million tokens per minute. Grow lifts Luna to 30,000 requests and 180 million tokens per minute. Five hundred dollars of lifetime spend buys you a production-grade ceiling, which is a short road.
Anthropic uses Start, Build and Scale tiers with monthly spend caps of $500, $1,000 and $200,000, and parks new organizations in an Evaluation tier with lower, unpublished limits until usage history builds. Start tier gives Sonnet 5.5 and Haiku 5.5 each 4,000 requests, 2 million input tokens and 400,000 output tokens per minute. Cached reads don’t count toward input tokens per minute on current models, so a heavily cached app gets far more real throughput than the number suggests.
Startup credits if you have a VC
OpenAI for Startups routes through venture partners. You need a unique referral code from your VC, and benefits may include free API credits, rate limit upgrades and time with solutions engineers. No credit amount is published, and once credits run out the account switches to standard billing. Without a partner investor there’s no door to knock on.
Claude Startups is between cohorts, with the current one full. The listed benefits are discounts and credits worth up to $45,000 from companies building on Claude, redeemed with a code in the Console, plus up to $100,000 in additional API credits for founders backed by a VC in Anthropic’s partner network. Those credits work only on the first-party API, not on AWS, Google Cloud or Azure, so a Bedrock shop gets nothing from them.
Prepaid billing and your data
Both platforms are prepaid for self-serve accounts. OpenAI’s tier thresholds are measured in credit purchases, and Anthropic’s Console sells credits that are non-refundable and expire one year from purchase, with optional auto-reload. Anthropic also pauses the API at the tier cap until the first of the next month unless you request an increase, and it returns a 429 without a retry-after header when that happens. Watch for that one in your error handling.
On data, OpenAI keeps abuse monitoring logs for up to 30 days and doesn’t train on API data by default, with Zero Data Retention available to approved customers. Anthropic deletes API inputs and outputs within 30 days, doesn’t train on commercial inputs or outputs by default, and offers a ZDR agreement. Flagged policy violations can be held longer on both sides.
Our call for most teams
With identical list prices, the choice turns on your traffic shape. A chatbot that rereads a long system prompt thousands of times a day gets the lower effective rate on Anthropic because of 0.05x cached reads and cache-aware rate limits. A pipeline of unique, one-shot prompts pays the same on either, and OpenAI spares you the cache write premium. We’d measure the cache hit rate for a week before committing a quarter of spend to either.
The winner: Tie
List prices are identical across all three model classes, both discount batch work by half, and neither offers a signup credit. Anthropic wins on cached reads and subscriber credits. OpenAI wins on simpler caching math and limits that are published from the first purchase. The better deal depends on how repetitive your prompts are, and nothing else.
Choose OpenAI if your prompts are mostly unique, you want rate limits published from your first $5 purchase, and you can use Flex processing for background jobs.
OpenAI API offers (5)Choose Anthropic if you cache long context, already pay for Claude Max or Team and want the monthly API credits, or can wait out the Evaluation tier to reach Scale.
Anthropic API offers (5)