*Fourth in a set with [Big Tech AI Subscriptions](/publications/big-tech-ai-subscriptions/),
[If the Big Labs Block You](/publications/if-the-big-labs-block-you/) and
[Chinese AI Services](/publications/chinese-ai-services/). Those three cover
**subscriptions** — a human logging into a chat window. This one covers
**tokens**: metered API credit, which is what an autonomous agent actually
consumes. Prices as listed in early October 2026, USD per million tokens
(input/output) unless noted. This tier moves monthly; verify before
committing.*

## Why a subscription is the wrong product

A $20 subscription buys a seat for a person. An agent has no seat. It has a
loop: read, call a tool, read the result, decide, call again — and every pass
re-sends the accumulated context. Three things follow, and they matter more
than the headline price.

**Token amplification is the real bill.** One user instruction becomes many
model calls — thinking, searching, tool-calling, summarising — and you pay for
every intermediate step. AWS documents this plainly for Bedrock Agents: there
is no per-invocation fee, and that is precisely the trap, because the cost
lands on the internal calls instead. Price the loop, not the prompt.

**Cached input and batch discounts outrank the sticker price.** An agent
re-sends the same system prompt and the same conversation prefix on every
turn, which is exactly the shape caching is built for. OpenAI reads cached
input at 10% of standard rates and halves both directions on the Batch API;
xAI charges $0.50 cached against $2.00 fresh on Grok 4.7; Moonshot charges
$0.30 cache-hit against $3.00 cache-miss on Kimi K3 — a 10× difference on the
input side alone. A loop that reuses its prefix pays a fraction of the table
rate; one that rebuilds the prompt each turn pays all of it.

**The spend tier binds before the price does.** Every major vendor gates new
accounts behind a monthly cap that an agent can hit in an afternoon. See
[the last section](#what-actually-stops-an-agent) — this is the part most
cost comparisons leave out, and the part that actually breaks things.

## Big tech

The same eight companies as [the subscriptions article](/publications/big-tech-ai-subscriptions/),
asked a different question: can you buy tokens, and at what rate?

| Company | Tokens for sale? | Flagship | Mid / cheap | Notes |
|---|---|---|---|---|
| **OpenAI** | Yes | [GPT-6 Astra](https://www.cloudzero.com/blog/openai-pricing/) **$10 / $50** | GPT-5.6 Terra **$2 / $12** · Luna **$0.20 / $1.20** | Astra's context is 272K, and beyond it input doubles and output rises 50%. The 5.6 family is 1.05M across all tiers. Cached input 10%, Batch 50% |
| **Anthropic** | Yes | [Claude Fable 5](https://platform.claude.com/docs/en/pricing) **$10 / $50** | Opus 5 **$5 / $25** · Sonnet 5 **$3 / $15** · Haiku 4.5 **$1 / $5** | 1M context on everything except Haiku (200K). Fable 5 requires 30-day data retention — not available under zero-retention terms |
| **Google** | Yes | [Gemini 3.1 Pro](https://www.cloudzero.com/blog/gemini-pricing/) **$2 / $12** | 3.8 Flash **$0.75 / $3.75** · Flash-Lite from **$0.10 / $0.40** | Pro doubles to **$4 / $18** above 200K prompt. **The Flash line goes to $1.50 / $7.50 on 1 January 2027** — a 2× rise already on the calendar |
| **Microsoft** | Mostly other people's | — | MAI-Image-2.5 **$5 / $47** · MAI-Voice-2 **$22**/M chars | Foundry is a marketplace first. Its own [MAI line](https://microsoft.ai/news/today-were-announcing-3-new-world-class-mai-models-available-in-foundry/) is speech, voice and image; the text model, [MAI-Thinking-1](https://ai.azure.com/catalog/models/MAI-Thinking-1) (1T total / 35B active), is in preview with **no published price** |
| **xAI** | Yes | [Grok 4.7](https://benchlm.ai/xai/api-pricing) **$2 / $6** | Grok 4.3 **$1.25 / $2.50** · grok-build-0.1 **$1 / $2** | At or above a 200K prompt, Grok 4.7 bills **the entire request** at $4 / $12. Grok 4.3 is the 1M-context option and a third the output price |
| **Meta** | Yes, new | [Muse Spark 1.1](https://developer.meta.com/ai/products/meta-model-api/) **$1.25 / $4.25** | Contributor tier **$0.10 / $0.20** | 1M context, built for tool use and computer use; $20 free credits. **US developers only at launch**, and the cheap tier's price is [paid in training data](https://aiweekly.co/alerts/meta-prices-muse-spark-11-api-at-125425-per-m-tokens) — Meta trains on your prompts and completions |
| **Amazon** | Yes | [Nova Premier](https://www.cloudroutehq.com/aws-ai/amazon-nova-pricing) **$2.50 / $12.50** | Nova Pro **$0.80 / $3.20** · Micro **$0.035 / $0.14** | Bedrock also resells almost everyone else, so it doubles as a gateway |
| **Apple** | **No** | — | — | No API, no tokens, nothing to buy |

**The spread inside one vendor is wider than the spread between vendors.**
OpenAI's own catalogue runs 25× on input between Luna and its top tiers;
Amazon's runs 70× between Micro and Premier. Choosing Google over OpenAI moves
your bill by a fraction of what choosing Flash over Pro does. Pick the tier
first, the vendor second.

Two smaller observations worth the ink. **Microsoft is the odd one out** — it
sells a storefront rather than a model, which is a fine thing to buy and a bad
thing to mistake for independence. And **Meta's Contributor tier is the only
place in this table where the price is explicitly data** rather than money;
that is a clean trade if the workload is non-sensitive and a disqualifying one
if it isn't.

## Alternatives from other countries

Same exclusions as the big-tech table above, and Chinese developers get
[their own section](#chinese-models). What's left is thinner than the
subscription market — but not in the way you'd expect, because **Cohere sells
tokens perfectly well.** It was excluded from
[the alternatives article](/publications/if-the-big-labs-block-you/) for having
no self-serve consumer subscription, and that disqualification is simply
irrelevant here: there is a published per-token price and a card form.

| Developer | Country | Models and rates | Notes |
|---|---|---|---|
| **Mistral AI** | 🇫🇷 France | [Large 3](https://www.cloudzero.com/blog/mistral-api-pricing/) **$0.50 / $1.50** · Medium 3.5 **$1.50 / $7.50** · Small 4 **$0.15 / $0.60** · Magistral Medium **$2 / $5** · Codestral **$0.30 / $0.90** | The strongest non-US, non-Chinese option by a distance. Runs on European infrastructure; Large 3 undercuts every Western flagship on output price |
| **Cohere** | 🇨🇦 Canada | Command A **$2.50 / $10** · Command R **$0.15 / $0.60** · Command R7B **$0.0375 / $0.15** | Enterprise-shaped but self-serve on tokens. R7B is among the cheapest credible models anywhere |
| **Upstage** | 🇰🇷 South Korea | [Solar Pro 4](https://openrouter.ai/upstage/solar-pro4) list **$0.30 / $1.20** (promo **$0.09 / $0.36**) · Solar Mini 4 **$0.05 / $0.20** | Mini 4 is a 35B MoE with 3B active and a **524K context** at five cents — the best context-per-dollar in this table. English, Korean and Japanese |
| **Aleph Alpha** | 🇩🇪 Germany | Luminous Supreme ~**$5** input | Priced as a compliance product, not a capability one: on-premise deployment, audit logging, data-isolation guarantees |
| **Sarvam AI** | 🇮🇳 India | Sarvam 105B / 30B listed **free** via hosts | Indic-language strength, no paid tier to escalate to, and nothing here is built for code |
| **OVHcloud · Scaleway · Nebius** | 🇫🇷 🇳🇱 EU | Open weights on EU metal; OVHcloud ~**€0.63–0.67**/M blended | Hosting, not training — you buy someone else's model under EU jurisdiction. [The GPU route](https://www.qovery.com/blog/migrate-gpu-inference-european-sovereign-cloud-playbook) is the other half: H100 at ~€2.80–2.87/hour against ~€6.03 at AWS and ~€10.78 at Azure |

**Where I could not find a token to buy.** Japan has a real model ecosystem
and no self-serve API shop I could confirm a price for — Preferred Networks
sells PLaMo access on enquiry, Sakana is geo-restricted. Israel's AI21 halted
Jamba development in May 2026, so the obvious entry is gone. For the UAE's
Falcon line I found downloadable weights but no first-party per-token price I
could verify; treat it as self-hosting. In each case the honest statement is
"not verified", not "doesn't exist" — but if you can't find the price page,
you can't plan a budget around it either.

The pattern underneath: **outside the US and China, the market sells you
jurisdiction more than capability.** Mistral is the one entry competing on
both. Everything else in that table is either a compliance posture, a language
specialism, or a European address wrapped around somebody else's open weights
— all legitimate reasons to buy, none of them a capability argument.

## Chinese models

This is where the cheap tokens are, and it isn't close.

| Service | Developer | Rates | Notes |
|---|---|---|---|
| **DeepSeek** | DeepSeek | [V4.1 Flash](https://benchlm.ai/deepseek/api-pricing) **$0.30 / $1.20**, **half off-peak** · V4 Pro **$1.32 / $3.96** | The off-peak discount is a genuine agent feature: batch an overnight run and the same work costs **$0.15 / $0.60** |
| **Qwen** | Alibaba | [Qwen3.5 Flash](https://benchlm.ai/alibaba/api-pricing) **$0.10 / $0.40** · Plus **$0.40 / $2.40** · 397B **$0.60 / $3.60** · Qwen3.7 Flash **$0.03 / $0.13** | Qwen3.7 Flash at three cents is the cheapest usable token I found anywhere in this article — roughly a seventh of Gemini Flash-Lite |
| **Kimi** | Moonshot AI | [K3](https://benchlm.ai/moonshot/api-pricing) **$3 / $15**, cache-hit input **$0.30** · K2.6 and K2.7 Code **$0.95 / $4.00** | K3 is priced like a Western flagship. The 10× cache-hit discount is what makes it viable in a loop |
| **GLM** | Z.ai | GLM-5.3 Flash **$0.15 / $0.50** · GLM-5.2 **$0.91 / $2.86** | Trackers disagree on GLM-5.2 — one lists **$1.40 / $4.40**. Confirm in-console before budgeting |
| **MiniMax** | MiniMax | M2.5 ~**$0.20**/M blended | |
| **Doubao** | ByteDance | ~**$0.12**/M input | The consumer app is China-only; the model is reachable through gateways and Volcano Engine |

Three things to weigh against those numbers, in descending order of how often
they actually matter.

1. **Latency compounds in a loop.** China-hosted endpoints add roughly
   [100–200 ms per call](https://www.edenai.co/post/beyond-us-ai-providers-asian-and-european-llm-apis-for-production)
   from Europe or the US. On a chat turn that's invisible. On a forty-step
   agent trace it's four to eight seconds of pure wire time, before the model
   thinks about anything.
2. **Jurisdiction is the same question as always.** Prompts and tool outputs
   land with a company subject to PRC law. An agent is worse than a chat
   window here, not better: it sends whatever it happens to read, including
   files you never consciously pasted.
3. **Sources disagree more than for Western vendors.** Regional pricing gates
   mean third-party trackers and the in-console figure often differ — the GLM
   row above is a live example, and DeepSeek V4 Pro is quoted anywhere between
   $0.44 and $1.32 depending on who you ask.

And the standing escape hatch from
[the Chinese services article](/publications/chinese-ai-services/) still
applies, more strongly for agents than for chat: **the weights are
downloadable.** Running DeepSeek or GLM on your own hardware removes points 1
and 2 completely, and an agent's token volume is exactly the profile where
owning the hardware starts to beat renting the tokens. See
[the multi-GPU survey](/publications/local-ai-models-2026/).

## Gateways: the fourth place to buy

Most agent frameworks don't call a vendor directly — they call a gateway, and
for failover that's the right instinct, since a single account outage stops an
autonomous run cold.

- **[OpenRouter](https://hostfleet.net/openrouter-vs-together-vs-groq-vs-fireworks/)**
  passes through vendor pricing and charges **5.5% on credit purchases**,
  routing to Fireworks, Together, Groq and Cerebras underneath. You pay a
  margin for one API key and one bill across every vendor in this article.
- **Groq and Cerebras** publish per-token rates with no purchase fee and run
  600–750 tokens/sec decode on 70B-class models — often 2–3× Together's price
  for the privilege. On Llama 3.1 8B: Groq **$0.05 / $0.08**, Cerebras
  **$0.10 / $0.10**, Together and Fireworks **$0.18** both directions.
- **Fireworks** positions explicitly for agentic workloads — structured
  output, function calling, JSON mode — which is a product claim rather than a
  price one, but the right claim for this use case.

For an agent, the speed difference is not cosmetic. A forty-step loop at 700
tokens/sec finishes while a 70-tokens/sec endpoint is still on step six.

## What actually stops an agent

Not the price. The cap. Every major vendor gates new accounts, and an
unattended loop reaches the ceiling far faster than a human ever would.

| Vendor | Entry | How you climb | Top standard cap |
|---|---|---|---|
| **OpenAI** | [$5 minimum](https://benchlm.ai/blog/posts/api-credits-explained) prepaid, Tier 1 capped at **$100/month** | Five tiers unlocked by cumulative spend ($50 → $1,000) | **$200,000/month** at Tier 5 |
| **Anthropic** | Start tier, **$500/month** | Start → Build → Scale, then a custom uncapped tier via the account team | **$200,000/month** at Scale |
| **Google** | Tier 1 **$250/month** on a linked billing account | $100 paid + 3 days → Tier 2 ($2,000); $1,000 paid + 30 days → Tier 3 | **$20,000–$100,000+** at Tier 3 |

Three practical consequences:

- **Credits expire after a year** at all three, and the expiry can't be
  extended. Prepaying a large balance to secure a rate is a bet on your own
  burn rate.
- **Auto-recharge is on by default at OpenAI**, with a $5 minimum top-up. An
  agent in a retry loop plus auto-recharge is an unbounded bill — set the
  optional monthly cap the same day you create the key.
- **Google is the only one of the three offering postpay**, which sidesteps
  both the prepayment and the expiry.

## My take

**Buy on tier, not on brand.** The honest conclusion from the big-tech table is
that the vendor choice barely moves the bill compared to the tier choice. A
sensible agent runs a cheap workhorse for the ninety percent of steps that are
routine — Gemini Flash-Lite at $0.10, Qwen3.5 Flash at $0.10, GPT-5.6 Luna at
$0.20 — and escalates to a frontier model only for the steps that need
judgement. Paying Astra or Fable 5 rates for "read this file and tell me the
next step" is the single most expensive habit in agent engineering.

**Then make the prefix cacheable.** The cached-input discounts in these tables
range from 4× to 10×, and an agent loop is the one workload shaped to collect
them in full. That is a larger saving than any vendor switch on this page, and
it is a code change rather than a procurement decision.

**Hold two accounts.** Not for price arbitrage — for the run that dies at 3am
because one tier cap was reached or one provider refused one prompt. A gateway
account at 5.5% is a reasonable insurance premium against a stalled
autonomous job; a second direct account is cheaper and more work.

**On the cheap Chinese tokens:** three cents a million is real, and so is the
jurisdiction. For public data, scraping, summarising, classification — the
bulk of what agents actually do — the price is hard to argue with. For
anything touching client data, the answer that keeps working is the one from
the companion article: take the weights and run them yourself. An agent's
volume is exactly what makes that arithmetic close.

The thread running through all four of these articles: **the subscription
market is crowded and the token market is concentrated.** Eleven of the
sixteen entries above are American or Chinese, and the non-US alternatives are
mostly selling a jurisdiction rather than a capability. If you want genuine
independence for an autonomous agent — the thing that keeps running when an
account, a country or a policy changes — the only complete answer is still
holding the weights.

## Further reading

- [OpenAI API pricing, every model](https://www.cloudzero.com/blog/openai-pricing/) · [Gemini API pricing and the January increase](https://www.cloudzero.com/blog/gemini-pricing/) · [Mistral API pricing](https://www.cloudzero.com/blog/mistral-api-pricing/)
- [Anthropic pricing](https://platform.claude.com/docs/en/pricing) — primary source
- [Amazon Bedrock pricing and token amplification](https://www.cloudzero.com/blog/amazon-bedrock-pricing/)
- [API credits, tiers and expiry compared](https://benchlm.ai/blog/posts/api-credits-explained) · [Cross-vendor token price table](https://benchlm.ai/llm-pricing)
- [Asian and European LLM APIs for production](https://www.edenai.co/post/beyond-us-ai-providers-asian-and-european-llm-apis-for-production) · [Inference API price comparison](https://infrabase.ai/compare/inference-apis)
- [Chinese AI Services](/publications/chinese-ai-services/) — the subscription side of the same labs
- [Self-Hosting the Frontier](/publications/local-ai-models-2026/) — the alternative to buying tokens at all