What My AI Agents Actually Cost: DeepSeek vs the Big APIs

In plain terms: A look at what running AI assistants around the clock actually costs. One month of heavy use came to about $24 on a budget AI service; the same workload would cost roughly $475 to $915 on the big-name flagship services. The trick that keeps it cheap is caching — most of the text sent to the AI is repeated, and repeated text is billed at a steep discount.

My home agent stack — eight named agents running 24/7 on DeepSeek's API — just finished its first full month. I pulled the usage exports and did the math: what did it really cost, and what would the identical workload cost on the big APIs?

tl;dr

June 2026, full month: $24.08 total ($9.54 flash + $14.54 pro). July 1–11 so far: $2.23.
Volume: ~757M tokens across ~13.6k requests (June + early July combined).
The kicker: ~87% of input tokens were cache hits — that ratio, not the sticker price, is what makes always-on agents affordable.
Same workload elsewhere (estimated): roughly $475–$915 on the flagships, $33–$230 on their cheap tiers.

The real numbers

Combined June + early-July volume: 659.8M cached-input tokens, 93.1M cache-miss input tokens, 4.4M output tokens — about 757M tokens total. The June slice alone (which the comparison below uses) was ~560.8M cache hits, ~86.2M cache misses, and ~3.8M output tokens, for $24.08.

That 87% cache-hit rate isn’t luck — it’s the natural shape of agent traffic. Every turn re-sends the same system prompt, tool definitions, and conversation history, and providers bill that repeated prefix at a tiny fraction of the normal input price. DeepSeek charges $0.0028–$0.003625 per million cached tokens; without caching, the same input volume would have cost ~35× more.

Same workload, other providers

Apples-to-apples estimate: June’s exact token mix priced on each provider’s current published rates (all verified 2026-07-11 — DeepSeek, Anthropic, OpenAI, xAI). Cache hits billed at each provider’s cached-input rate; misses at the normal input rate (Anthropic charges a 1.25× cache-write premium on misses, which is included).

Provider / model	Cached in $/M	Input $/M	Output $/M	June workload
DeepSeek v4 flash+pro (actual)	$0.0028–0.0036	$0.14–0.44	$0.28–0.87	$24.08
OpenAI GPT-5.4 nano	$0.02	$0.20	$1.25	~$33
Anthropic Claude Haiku 4.5	$0.10	$1.00 (+write)	$5.00	~$185
xAI Grok 4.3	$0.20	$1.25	$2.50	~$230
xAI Grok 4.5 (flagship)	$0.50	$2.00	$6.00	~$475
OpenAI GPT-5.6 Sol (flagship)	$0.50	$5.00	$30.00	~$825
Anthropic Claude Opus 4.8 (flagship)	$0.50	$5.00 (+write)	$25.00	~$915

Assumptions: numbers rounded to the nearest ~$5; June mix only (560.8M hit / 86.2M miss / 3.8M out); assumes each provider would achieve the same cache-hit ratio (realistic — the traffic shape is identical); ignores batch discounts and time-of-day pricing.

The honest part: quality differs

This is not “DeepSeek equals Claude for 4% of the price.” The frontier models — Claude Opus, GPT-5.6, Grok 4.5 — are noticeably stronger on complex, long-horizon, agentic work. But routine agent chatter, message triage, drafts, scheduled jobs, and glue automation don’t need frontier intelligence, and that’s ~95% of what an always-on stack does. That’s exactly the split my setup exploits: cheap tokens for the busywork, frontier brains for the hard stuff — the same division of labor from when the agent stack broke and Claude Code fixed it.

The frontier half of my usage runs through Claude Code on the Max subscription — a flat monthly fee, so the heavy coding and debugging work never touches per-token billing at all. The DeepSeek dollars cover only the 24/7 agents.

The other bill: Claude Code on Max

For completeness, the frontier side, measured from this box’s local session history: June 2026: 412 Claude Code sessions across 13 active days. July 1–11: 297 sessions across 8 active days — about 709 sessions and ~314 MB of transcripts in six weeks. These are long, tool-heavy agentic coding runs, not one-off chats.

All of it is covered flat by the Claude Max subscription at $200/month. Subscription usage isn’t metered, so there’s no exact API-equivalent figure — but hundreds of long Opus-class agent sessions at list rates ($5/$25 per million tokens) would plausibly land in the four figures monthly. The honest total for the whole stack:

~$200/mo flat — Max subscription, all the frontier coding and debugging
~$25/mo metered — DeepSeek, the 24/7 agents

Gotchas

Protect your cache-hit ratio. A timestamp or random ID near the top of a system prompt invalidates the cached prefix on every request — at these ratios that’s a 10–30× bill multiplier.
Output tokens barely matter here. 4.4M output vs 753M input; agents read far more than they write.
Prices move fast. Grok 4.5 launched three days before this post; recheck the pricing pages before planning around any number above.