DeepSeek · DeepSeek V3.2 · Reasoning
DeepSeek V3.2 (reasoner) pricing and limits
DeepSeek V3.2 in thinking mode. Same list price as the chat mode, but the chain of thought is billed as output, so output tokens per request are much higher.
| Line | Tokens | Rate / 1M | Cost |
|---|---|---|---|
| Input, uncached | 3,000 | $0.28 | $0.00084 |
| Input, cache read | 3,000 | $0.028 | $0.00008 |
| Output | 6,000 | $0.42 | $0.00252 |
- Input / 1M
- $0.28
- Output / 1M
- $0.42
- Cached / 1M
- $0.028
- Cache write / 1M
- n/a
- Blended 3:1
- $0.32
- Context
- 128K
What drives DeepSeek V3.2 (reasoner) cost
The reasoner returns its chain of thought in a separate reasoning_content field and bills those tokens as output. A response with a short final answer can carry thousands of reasoning tokens, so cost per request is several times the chat mode at the same rates.
Context caching works as in chat mode with cache hits at a tenth of input. Because reasoning output dominates, caching saves less in relative terms than on the chat model.
Cost of a typical call
Take a reasoning request with 6,000 input tokens, 3,000 of them served from cache, and 6,000 output tokens. At DeepSeek V3.2 (reasoner) list prices ($0.28 input, $0.028 cached input, $0.42 output per 1M) one call costs $0.00344, against $0.00420 with no cache hits. At 200,000 calls a month that is $689, or $840 uncached.
Blended at a 3:1 input to output ratio, DeepSeek V3.2 (reasoner) costs $0.32 per 1M tokens. Routing the same call to the fallback in the policy below, DeepSeek V3.2 (chat), costs $0.00344, which is what a tenant pays you in COGS once they cross the hard cap and PALM SOUQ downgrades instead of blocking.
Cost comparison
DeepSeek V3.2 (reasoner) against three alternatives
The same reasoning request priced on o4-mini, DeepSeek V3.2 (chat), Gemini 2.5 Flash. Move the sliders to match your traffic; models without a cache-read rate bill every input token in full.
Benchmark quality per task before switching. A cheaper model that needs more steps or longer outputs can cost more per finished task than its rate suggests, which is why PALM SOUQ reports cost per feature, not only per token.
DeepSeek V3.2 (reasoner) this guide
$0.00344 per call
$861 / mo
baseline
o4-mini
$0.0305 per call
$7,631 / mo
+786% vs DeepSeek V3.2 (reasoner)
DeepSeek V3.2 (chat)
$0.00344 per call
$861 / mo
+0% vs DeepSeek V3.2 (reasoner)
Gemini 2.5 Flash
$0.0161 per call
$4,031 / mo
+368% vs DeepSeek V3.2 (reasoner)
List prices from the PALM SOUQ LLM Price Index, computed per call as uncached input + cache reads + output. Reasoning tokens count as output on reasoning models.
Rate limits by usage tier
No fixed published rate limits; capacity is allocated dynamically and long reasoning requests can take significant time under load. Plan for long tail latency in your timeouts.
Exact limits vary by account tier and change over time, so we do not reprint them here. Check the limits page for your own account in the DeepSeek documentation at api-docs.deepseek.com. PALM SOUQ enforces your own per-tenant limits in front of the provider's, so a single customer cannot consume the shared allowance that every other tenant depends on.
Circuit-breaker best practices
Cap max output tokens per request, which bounds reasoning plus answer, and enforce a per-tenant reasoning token ceiling at the gateway.
Do not feed reasoning_content back into the next turn's prompt. It inflates input on every step and can create self-reflection spirals that the semantic echo check will then have to stop.
When a breaker trips, the gateway answers in under 50 ms with HTTP 429 and X-PalmSouq-Reason: circuit_breaker_triggered, before the request reaches DeepSeek. Your agent framework sees a normal rate-limit response and stops, and the audit log records which heuristic fired and for which tenant.
# palmsouq.policy.yaml
policy: deepseek-v3-2-reasoner-pro
match:
model: deepseek-v3.2-reasoner
tier: pro # from X-PalmSouq-Tier
budgets:
monthly_usd: 80 # per tenant
soft_cap_pct: 80 # webhook to your app
on_hard_cap: fallback # or: block (HTTP 429)
fallback:
- deepseek-v3.2-chat
breakers:
tool_thrashing: { identical_calls: 3, window_steps: 8 }
token_velocity: { max_tokens_per_s: 50000, window_s: 10 }
semantic_echo: { repeats: 3 }
max_steps: 12
max_reasoning_tokens: 16000 # per request
cache:
meter_reads: true- budgetsPer-tenant monthly dollar cap with an 80% soft-cap webhook, keyed by X-PalmSouq-Tenant.
- fallbackAt the hard cap, requests reroute to DeepSeek V3.2 (chat) ($0.32 blended per 1M) instead of failing.
- breakersAll four loop heuristics on, evaluated in the request path with under 5 ms added latency.
Where it fits
Math, logic and coding tasks where reasoning improves accuracy and price matters more than latency.
Provider documentation: api-docs.deepseek.com
DeepSeek V3.2 (reasoner) pricing questions
How much does DeepSeek V3.2 (reasoner) cost per million tokens?
At list price DeepSeek V3.2 (reasoner) costs $0.28 per 1M input tokens and $0.42 per 1M output tokens, with cached input at $0.028 per 1M. Prices were last reviewed in the PALM SOUQ LLM Price Index and are for the standard tier and shortest context band.
Does DeepSeek V3.2 (reasoner) support prompt caching?
Yes. Cached input is billed at $0.028 per 1M tokens against $0.28 for uncached input. PALM SOUQ records cached tokens per tenant from the provider's usage data, so cache savings show up in each customer's COGS.
What are the DeepSeek V3.2 (reasoner) rate limits?
No fixed published rate limits; capacity is allocated dynamically and long reasoning requests can take significant time under load. Plan for long tail latency in your timeouts. For your account's exact numbers, see the DeepSeek documentation at api-docs.deepseek.com.
How do I stop runaway DeepSeek V3.2 (reasoner) agent loops?
Cap max output tokens per request, which bounds reasoning plus answer, and enforce a per-tenant reasoning token ceiling at the gateway. The sample policy on this page enables all four PALM SOUQ heuristics with a max step cap of 12 and falls back to DeepSeek V3.2 (chat) at the hard cap.
Is there a batch discount for DeepSeek V3.2 (reasoner)?
No batch tier; DeepSeek discounts through automatic context caching instead. Use caching and model routing to lower cost for asynchronous jobs.
Govern DeepSeek V3.2 (reasoner) spend per customer.
Point your DeepSeek client at gateway.palmsouqltd.com, pass X-PalmSouq-Tenant, and every DeepSeek V3.2 (reasoner) token is attributed, capped and protected by circuit breakers.
14-day free production trial · 100k requests included · no card required