API model string
qwen3-max / qwen3-max-2026-01-23Context window
262K tokens
Max output 66K tokens
- Knowledge cutoff
- not published
- Released
- 2026-01-23 (snapshot)
- Modalities
- text in, text out
Last verified July 10, 2026
Rate limits by tier
DashScope does not publish named rate-limit tiers for Qwen3-Max in English documentation. Access levels are determined by account type and balance; exact RPM/TPM limits are visible only in the DashScope console under your account quota page.
| Tier | Requirements | RPM | TPM | RPD | Concurrent | Notes |
|---|---|---|---|---|---|---|
| Free trial (new accounts) | New Alibaba Cloud international account; Singapore endpoint only | not published | not published | not published | not published | 1M tokens per eligible model, 90-day expiry. Old free OAuth tier discontinued April 15, 2026 — this is a one-time onboarding credit only, not a recurring free plan. |
| Pay-as-you-go (standard) | USD balance on DashScope international account | not published | not published | not published | not published | No formal named tiers published; rate limits for Qwen3-Max specifically are not documented in English. Check DashScope console for your active quota. |
| Qwen Coding Plan Lite | ~$10/month subscription | not published | not published | not published | not published | Fixed monthly plan; specific Qwen3-Max RPM/TPM limits under this plan are not published. |
| Qwen Coding Plan Pro | ~$50/month subscription | not published | not published | not published | not published | Higher capacity than Lite; exact RPM/TPM not published. |
| Enterprise / Beijing endpoint | Alibaba Cloud sales contact | not published | not published | not published | not published | Beijing endpoint is 60-70% cheaper than Singapore but stores data in China — compliance review required. Custom limits and SLA available; specific numbers not published. |
Swipe the table sideways to see every limit column.
- 1.DashScope uses OpenAI-compatible endpoints; exact 429 error body strings are not prominently documented in English.
- 2.Batch API cannot be combined with context caching — they are mutually exclusive features.
- 3.Thinking-mode tokens are billed at the output rate ($6.00/MTok), which can significantly inflate costs on complex tasks.
- 4.Newer flagship Qwen3.7-Max (released 2026) is available at ~$1.25/$3.75 on 50% promotional pricing (list $2.50/$7.50). Qwen3-Max is prior-gen but actively served. Third-party prices vary widely: OpenRouter ~$0.78/$3.90; CloudPrice from $0.359/$1.43 vs DashScope Singapore $1.20/$6.00.
Limits verified against the Alibaba (DashScope) docs, July 10, 2026.
Token pricing
What you pay per million tokens (USD). Input and output are billed separately.
Input
$1.20
per 1M tokens
Output
$6.00
per 1M tokens
Batch discount
−50%
on batch jobs
- Prices are for DashScope international (Singapore) endpoint, updated July 1-2, 2026.
- Context caching is available but the cache-read rate is not published separately; calculate costs without a cached discount until DashScope publishes the rate.
- Batch API discount (50% off input + output) cannot be combined with context caching.
- Beijing endpoint is 60-70% cheaper but data is stored in China — verify compliance implications before switching.
- Thinking tokens (when thinking mode is enabled) bill at the output rate of $6.00/MTok.
- Third-party aggregators: OpenRouter ~$0.78 in / $3.90 out; CloudPrice.net from $0.359/$1.43. These rates offer significant savings but carry SLA and latency tradeoffs — verify before relying on them for production.
Side-project chatbot
$12.00
per month
Assumptions
5M input / 1M output tokens per month, DashScope $1.20/$6.00
5 × $1.20 + 1 × $6.00 = $6.00 + $6.00 = $12.00
Production app
$144.00
per month
Assumptions
60M input / 12M output per month; cache rate not published, no cached discount applied
60 × $1.20 + 12 × $6.00 = $72 + $72 = $144.00 (cache read rate not published — no reduction applied)
High-volume pipeline
$960.00
per month
Assumptions
400M input / 80M output per month, DashScope $1.20/$6.00
400 × $1.20 + 80 × $6.00 = $480 + $480 = $960.00; with batch 50% off: $480.00
Run your own numbers
Drag your real monthly token volumes and watch the bill update live — priced against rival models at the same usage.
Estimated qwen3-max / qwen3-max-2026-01-23 spend
$66/mo
Input: $36
Output: $30
30M in × $1.20 + 5M out × $6.00 = $66
Same volume, priced across models
- Kimi K2.6Cheapest$49
- qwen3-max / qwen3-max-2026-01-23This model$66
- Mistral Large 3$90
Rivals priced at their published input/output rates for the same monthly volumes. Prompt caching is model-specific, so it is applied to qwen3-max / qwen3-max-2026-01-23 only. Estimates for comparison; real bills vary with request shape and long-context surcharges.
qwen3-max / qwen3-max-2026-01-23 vs the alternatives
Qwen3-Max vs Kimi K2.6, DeepSeek V4, and Mistral Large 3 across pricing, limits, and features relevant to high-throughput production use.
| Aspect | qwen3-max / qwen3-max-2026-01-23 | Kimi K2.6 | DeepSeek V4 | Mistral Large 3 |
|---|---|---|---|---|
| Context window | 262K | 262K (Kimi K2.6) | not published (DeepSeek V4) | 128K (Mistral Large 3) |
| Input $/MTok (primary host) | $1.20 (DashScope) | $0.95 (Kimi K2.6) | n/a (not in manifest) | $2.00 (Mistral Large 3 est.) |
| Output $/MTok | $6.00 (DashScope) | $4.00 (Kimi K2.6) | n/a | $6.00 (Mistral Large 3 est.) |
| Batch discount | 50% | ~40% (Kimi) | n/a | 50% (Mistral Large 3) |
| Free trial tokens | 1M (onboarding, 90d) | none | none | yes (Experiment tier) |
| Thinking mode | yes (billed as output) | yes | yes | yes |
| Data residency | Singapore or China (separate endpoints) | China | not published | EU (Mistral Paris) |
| Pricing stability | Prior-gen, prices stable | EOL'd legacy K2 | volatile | stable |
Swipe the table sideways to see every model.
Hitting a 429? The playbook
The exact errors you'll see
HTTP 429 Too Many Requests (DashScope OpenAI-compatible endpoint){"error": {"code": "RateLimitExceeded", "message": "..."}} (expected OpenAI-compat pattern — exact DashScope body not prominently published in English docs; verify in your console)Retry-After: <seconds>Why it happens & how to fix it
Exceeded per-minute quota on international account
Reduce request cadence; open DashScope console to check your model's active RPM quota under Account > Quota Management.
No active USD balance on pay-as-you-go account
Add USD balance to your DashScope international account at dashscope.aliyuncs.com/recharge.
Batch job exceeding concurrent limits
Stagger batch submissions; do not combine batch requests with context caching (they are mutually exclusive).
Request too large for the current plan tier
Reduce prompt size or upgrade from pay-as-you-go to a Qwen Coding Plan subscription for higher capacity.
Retry strategy
DashScope uses OpenAI-compatible endpoints — standard HTTP 429 handling applies. Honor any Retry-After header if present. Use exponential backoff with jitter: start at 1s, double each attempt up to a 60s cap, add random jitter (0-1s) to prevent thundering herd. After backoff, ramp request rate gradually rather than returning immediately to full throughput.
1import OpenAI from 'openai';23const client = new OpenAI({4 apiKey: process.env.DASHSCOPE_API_KEY ?? '',5 baseURL: 'https://dashscope-intl.aliyuncs.com/compatible-mode/v1',6});78async function callWithRetry(9 prompt: string,10 maxAttempts = 611): Promise<string> {12 for (let attempt = 0; attempt < maxAttempts; attempt++) {13 try {14 const response = await client.chat.completions.create({15 model: 'qwen3-max-2026-01-23',16 messages: [{ role: 'user', content: prompt }],17 });18 return response.choices[0].message.content ?? '';19 } catch (err: any) {20 if (err?.status === 429 && attempt < maxAttempts - 1) {21 const retryAfter = err.headers?.['retry-after'];22 const wait = retryAfter23 ? parseInt(retryAfter, 10) * 100024 : Math.min(60000, 1000 * Math.pow(2, attempt)) + Math.random() * 1000;25 console.warn(`429 on attempt ${attempt + 1}; waiting ${wait}ms`);26 await new Promise((r) => setTimeout(r, wait));27 } else {28 throw err;29 }30 }31 }32 throw new Error('Max retry attempts reached');33}How to raise your limits
The ladder from the starter tier to enterprise — what each rung takes, and what it unlocks.
Onboarding trial
ImmediateSign up for Alibaba Cloud international (Singapore) account at alibabacloud.com
Unlocks: 1M free tokens per eligible model (dozens of models), valid 90 days
Pay-as-you-go
Immediate on depositAdd USD balance to DashScope international account
Unlocks: Full pay-per-token access at $1.20/$6.00; batch at 50% off
Qwen Coding Plan Lite
ImmediateSubscribe for ~$10/month via DashScope console
Unlocks: Fixed monthly capacity (exact RPM/TPM not published — visible in console after subscribing)
Qwen Coding Plan Pro
ImmediateSubscribe for ~$50/month via DashScope console
Unlocks: Higher fixed capacity than Lite (exact RPM/TPM not published)
Beijing endpoint
ImmediateSwitch to Beijing regional endpoint in your API base URL
Unlocks: 60-70% lower prices vs Singapore — confirm data-residency compliance before switching (data stored in China)
Enterprise
Sales cycleContact Alibaba Cloud sales via alibabacloud.com/contact-us
Unlocks: Custom limits, SLA, dedicated quota, advanced support
Cut your token spend
Switch to Beijing endpoint if data residency allows
60-70% cost reduction vs Singapore endpointUpdate your base URL to the Beijing DashScope endpoint; verify compliance with your data governance policy before switching — data is stored in China.
Batch API for async workloads
50% off input + output tokensUse DashScope batch API for non-real-time jobs. Note: batch cannot be combined with context caching — choose one or the other per workload type.
Context caching for repeated content
Cost savings on cache hits (rate not published yet — monitor DashScope docs)Enable context caching for long system prompts or static documents you reuse across requests. Cache read price has not been published separately as of July 10, 2026 — track DashScope pricing page for the rate before budgeting.
Disable thinking mode for simple tasks
Eliminate thinking-token costs billed at $6.00/MTok output rateOnly enable thinking mode for tasks requiring deep reasoning (complex analysis, multi-step math). For classification, extraction, or simple generation, disable thinking to avoid significant output-token inflation.
Third-party aggregators for cost arbitrage
~35% savings on input (OpenRouter ~$0.78 vs DashScope $1.20)OpenRouter and CloudPrice serve Qwen3-Max at meaningfully lower rates (per third-party trackers — confirm on each platform before relying on these figures). Verify SLA, latency, and uptime tradeoffs before routing production traffic.
Upgrade to Qwen3.7-Max for new projects
Better capability at similar or lower price during 50% promo ($1.25/$3.75)Qwen3.7-Max is the current DashScope flagship. If starting a new integration, prefer qwen3.7-max — Qwen3-Max is prior-gen and will eventually be phased out.
Pin versioned API string
Prevents silent model drift and behavior changesUse qwen3-max-2026-01-23 (the snapshot alias) instead of the bare qwen3-max alias to lock to a specific checkpoint version.
Frequently asked questions
How do I increase Qwen3-Max rate limits?
DashScope does not publish named rate-limit tiers for Qwen3-Max. Your quota is visible in the DashScope console under Account > Quota Management. To raise limits, subscribe to a Qwen Coding Plan (Lite ~$10/mo or Pro ~$50/mo) or contact Alibaba Cloud sales for enterprise quotas.
Is Qwen3-Max API free?
New Alibaba Cloud international accounts receive 1M free trial tokens per eligible model, valid 90 days. After that, access requires a USD balance on pay-as-you-go or a Qwen Coding Plan subscription. The old free OAuth tier was discontinued April 15, 2026.
What does the Qwen3-Max 429 error mean?
A 429 on DashScope means you have exceeded your account's rate limit. DashScope does not prominently publish the exact error body in English — expect a JSON response similar to {"error": {"code": "RateLimitExceeded"}}. Add exponential backoff with jitter and check your quota in the DashScope console.
Qwen3-Max vs Qwen3.7-Max: which should I use?
Qwen3.7-Max is the newer DashScope flagship. If you are starting a new project, prefer Qwen3.7-Max (~$1.25/$3.75 on 50% promotional pricing as of July 2026, vs Qwen3-Max at $1.20/$6.00). Qwen3-Max is still actively served and is a valid choice if you have an existing integration — just pin the versioned string qwen3-max-2026-01-23.
Can I use Qwen3-Max cheaper on OpenRouter?
Yes — per third-party trackers, Qwen3-Max is available on OpenRouter at approximately $0.78 input / $3.90 output per MTok, versus DashScope's $1.20/$6.00. Verify current rates on OpenRouter before routing production traffic; SLA and latency differ from DashScope directly.
Does Qwen3-Max support context caching?
Context caching is listed as an available feature on DashScope, but the cache-read price has not been published separately as of July 10, 2026. Monitor the DashScope pricing page. Also note: Batch API (50% off) and context caching are mutually exclusive — you cannot combine them on the same request.
How do I set up Qwen3-Max with the OpenAI SDK?
Point the OpenAI SDK at DashScope's compatible endpoint: set base_url to https://dashscope-intl.aliyuncs.com/compatible-mode/v1 and use your DASHSCOPE_API_KEY as the API key. Use model string qwen3-max or the pinned snapshot qwen3-max-2026-01-23. The API interface is otherwise identical to OpenAI's chat completions.
Is Qwen3-Max suitable for teams with strict data residency requirements?
The Singapore endpoint keeps data outside China and is the recommended path for international teams with compliance requirements. A Beijing endpoint exists at 60-70% lower prices but stores data in China — run a formal compliance review before switching. Contact RapidDev (rapidevelopers.com/contact) if you need help architecting a multi-region AI pipeline that satisfies your governance requirements.
What are Qwen3-Max thinking mode costs?
Thinking mode is opt-in on Qwen3-Max. When enabled, the reasoning tokens are billed at the output rate of $6.00 per MTok. On complex multi-step tasks the thinking tokens can equal or exceed the regular output, effectively doubling costs. Disable thinking mode for tasks that do not require deep reasoning (classification, extraction, summarization).
We build AI apps that don't hit rate limits
- Retry, backoff & caching built in
- Multi-provider fallback routing
- Fixed price, you own the code
30-min call. No commitment.