TL;DR Qwen3.8 Max Preview is available for early testing through Alibaba's Token Plan ecosystem and supported Qoder products, but standard Model Studio pay-as-you-go API pricing has not yet been published. For production workloads, Qwen3.7 Max remains the safer baseline while Qwen3.8 is better suited for controlled evaluation.
Qwen3.8 Max Preview is the latest flagship preview associated with Alibaba's Qwen3.8 generation. Developers can already test it through supported products such as Qwen Code and Qoder, while key production detailsโincluding standard API pricing, rate limits, regional availability, and stable versioningโremain incomplete.
For teams already using Qwen3.7 Max, the key questions are straightforward: Is the Qwen3.8 Max API ready for production, how much does it cost, and is it worth migrating?
This guide focuses on those decisions, including current access methods, Qoder Credit pricing, confirmed specifications, and how to evaluate Qwen3.8 Max Preview against Qwen3.7 Max.
Qwen's official Qwen3.8 teaser highlights the 2.4T parameter count and planned open-weight release. The hosted qwen3.8-max-preview is already available through selected products, while the broader open-weight release is still upcoming.
Source*:* Qwen on X
Qwen3.8 Max API pricing
As of July 20, 2026, Alibaba has not published a standard price per million input or output tokens for Qwen3.8 Max.
Developers should therefore avoid assuming that Qwen3.8 will reuse Qwen3.7 Max pricing, that Qoder Credits equal direct API token costs, or that the qwen3.8-max-preview model ID will remain unchanged after general availability.
Qwen3.8 Max Preview at a Glance
| Item | Status as of July 20, 2026 |
|---|---|
| Model ID | qwen3.8-max-preview |
| Release stage | Preview |
| Reported total parameters | 2.4 trillion |
| Current access | Alibaba Token Plan ecosystem and supported Qoder products |
| Standard Model Studio pricing | Not yet published |
| Qoder standard Credit rate | 0.5x |
| Qoder promotional rate | 0.05x regular hours; 0.01x off-peak |
| Context window | 1,000,000 tokens via Qwen Code integration metadata |
| Thinking mode | Enabled in Qwen Code integration metadata |
| Image and video input | Enabled in Qwen Code integration metadata |
| Open weights | Announced as coming soon |
| Best use today | Controlled evaluation |
The key distinction is between Qwen3.8, the newly announced model generation, and qwen3.8-max-preview, the preview route currently appearing in supported products.
Qwen's official announcement describes Qwen3.8 as a 2.4-trillion-parameter model and says an open-weight release is planned. Qwen Code separately added qwen3.8-max-preview to its Token Plan integration.
What Is Confirmed About Qwen3.8 Max Preview?
Because Qwen3.8 is still in preview, it is useful to separate official announcements from integration metadata.
| Claim | Evidence level | Current status |
|---|---|---|
| 2.4T total parameters | Official announcement | Reported by the Qwen team |
| qwen3.8-max-preview model ID | Official integration | Added to Qwen Code Token Plan |
| 1,000,000-token context | Integration metadata | Configured in Qwen Code |
| Thinking mode | Integration metadata | Enabled in Qwen Code |
| Image and video input | Integration metadata | Enabled in Qwen Code |
| Open-weight release | Official intent | Announced as coming soon |
| Standard API token pricing | Unknown | Not yet publicly listed |
| Architecture and active parameters | Unknown | Not yet publicly documented |
| Independent benchmark consensus | Limited | Too early for a stable conclusion |
The Qwen Code configuration provides useful evidence for the 1M context window, thinking mode, and multimodal input support. However, these remain integration-level specifications until Alibaba publishes a complete Model Studio model card.
What does "Thinking mode" mean here? Qwen Code enables thinking behavior for
qwen3.8-max-preview. In the broader API ecosystem, models with this capability are often described as reasoning models, where additional computation may be used before generating a final response. Final accounting and pricing for reasoning or thinking tokens have not yet been documented for a standard Qwen3.8 Max API.
How to Access Qwen3.8 Max Preview
Confirmed access paths currently include:
- Alibaba's Token Plan ecosystem
- Qwen Code
- Qoder Desktop
- Qoder JetBrains Plugin
- Qoder CLI
- Qoder Cloud Agents
- Qoder Web and Mobile products
Qwen Code's official repository includes qwen3.8-max-preview in its Token Plan integration, while Qoder's launch documentation lists the model across supported Qoder products.
Developers building direct API integrations should confirm the exact model ID, endpoint compatibility, regional availability, rate limits, context limits, and billing behavior before using the preview in production systems.
Qoder Qwen3.8 Pricing: What Do 0.05x and 0.01x Mean?
Qoder is currently offering a promotional discount for Qwen3.8 Max Preview.
| Period | Qwen3.8 Max Preview Credit rate | Discount vs. 0.5x standard rate |
|---|---|---|
| Regular hours | 0.05x | 90% off |
| Off-peak hours | 0.01x | 98% off |
The promotion began on July 19, 2026, with the end date currently listed as TBD.
Qoder defines its off-peak period as 22:00โ08:00 Singapore Time (UTC+8). During US daylight-saving time in July 2026, this corresponds approximately to:
- 10:00 AMโ8:00 PM EDT
- 7:00 AMโ5:00 PM PDT
Qoder Credits are not API token prices
Qoder Credits is a product-level resource unit. Consumption can vary according to the model, token usage, and the number of calls triggered by agent or sub-agent workflows.
This calculation is valid:
discounted workflow credits =
base workflow credits ร applicable Qoder multiplier
But this is not:
Qwen3.8 API token price =
0.05 ร an assumed price per million tokens
For teams evaluating the model through Qoder, a more useful metric is:
credits per successful task =
total Credits consumed / successful tasks
This reflects the actual cost of completing a workflow without incorrectly converting Credits into API token prices.
Qwen3.8 Max Preview vs Qwen3.7 Max
For developers already using Qwen3.7 Max, the practical question is whether the preview is ready to replace it.
| Dimension | Qwen3.7 Max | Qwen3.8 Max Preview |
|---|---|---|
| Release status | Established production model | Preview |
| Standard pay-as-you-go pricing | Published | Not yet published |
| Regional deployment | Documented | Not yet fully documented |
| Versioned model IDs | Available | Not yet documented for GA |
| Context caching | Documented | Production rules pending |
| Batch support | Available on supported routes | Not yet documented |
| Cost modeling | Predictable | Limited |
| Best use today | Production | Evaluation |
Alibaba Cloud documents Qwen3.7 Max pricing across different deployment scopes. Prices and promotions may vary by region, so developers should check the official Model Studio pricing page.
For teams using CometAPI, the Qwen3.7 Max API provides a practical production baseline for evaluating future Qwen3.8 workloads.
For lighter or more cost-sensitive use cases, Qwen3.7 Plus provides another comparison point.
Should Qwen3.7 Max Users Upgrade to Qwen3.8 Now?
The practical approach is:
Test Qwen3.8 Max Preview now, but migrate only when its quality gains justify the operational and cost trade-offs.
Test Qwen3.8 Max Preview if:
- Your workload includes difficult coding or agent tasks.
- You are evaluating reasoning-heavy workflows.
- You need long-context or multimodal capabilities.
- You can tolerate preview-stage behavior changes.
- You have automated tests or human-review rubrics.
Keep Qwen3.7 Max in production if:
- You need predictable token costs.
- Regional deployment matters.
- Stable model IDs and versioning are required.
- Your architecture depends on documented caching or batch behavior.
- Your existing Qwen3.7 Max performance already meets requirements.
The reported 2.4T parameter count should not determine migration on its own.
For production AI systems, solve rate, latency, retries, human correction time, and cost per successful task are usually more useful than total parameter count.
How to Evaluate Qwen3.8 Max Preview
Rather than relying on launch claims or a single benchmark, build a small evaluation set based on your actual workloads.
A practical starting point is 20โ30 representative tasks.
| Workload | Example tasks | What to measure |
|---|---|---|
| Repository coding | Multi-file fixes, refactoring, tests | Pass rate, retries, tool calls |
| Full-stack development | Feature implementation | Completion quality, correction time |
| Data analysis | CSVs, tables, scripts | Accuracy, reasoning, artifact quality |
| Office workflows | Documents and spreadsheets | Completeness, human edits |
| Long-context tasks | Large repositories or documents | Recall, consistency |
| Reasoning tasks | Multi-step technical problems | Accuracy, latency |
| Routine coding | Simple fixes | Whether the larger model adds value |
For each task, record:
- Pass or fail
- Model and tool calls
- Retries
- End-to-end latency
- Input and output tokens
- Reasoning tokens where exposed
- Qoder Credits consumed
- Human correction time
For Qwen3.7 Max:
cost per successful task =
total API cost / successful tasks
For Qwen3.8 Max Preview through Qoder:
credits per successful task =
total Credits consumed / successful tasks
Once standard Qwen3.8 API pricing becomes available, rerun the same evaluation and compare:
quality ร latency ร reliability ร cost
That provides a more useful migration signal than benchmark scores alone.
What Developers Should Watch Next
The most important Qwen3.8 updates to watch are:
- Official Model Studio model card.
- Input, output, cached-input, and reasoning-token pricing.
- Regional API availability and rate limits.
- Stable model IDs and versioning.
- Context Cache and Batch support.
- Open-weight repository and license.
- Architecture and active-parameter details.
- Independent coding, agent, multimodal, latency, and efficiency benchmarks.
FAQ
Is the Qwen3.8 Max API available?
Qwen3.8 Max Preview is accessible through Alibaba's Token Plan ecosystem and supported Qoder products. A complete general-availability Model Studio API offering has not yet been publicly documented.
How much does the Qwen3.8 Max API cost?
Standard per-million-token pricing has not yet been published. Qoder currently offers promotional Credit rates of 0.05x and 0.01x, but these are Qoder Credits rather than general API token prices.
Does Qwen3.8 Max Preview have a 1M context window?
Qwen Code's Token Plan integration metadata configures a 1,000,000-token context window for qwen3.8-max-preview. Endpoint-specific production limits should still be confirmed when Alibaba publishes final documentation.
Can I use Qwen3.8 Max through an OpenAI-compatible API?
A broadly documented production Qwen3.8 Max endpoint with a final OpenAI-compatible API contract has not yet been published. Developers should verify the exact preview endpoint and supported parameters before integration.
Should I replace Qwen3.7 Max with Qwen3.8 Max Preview?
For most production systems, not yet. Use Qwen3.7 Max as the stable baseline and evaluate Qwen3.8 Max Preview until its pricing and production behavior are better documented.
Prepare for Qwen3.8 on CometAPI
Qwen3.8 Max support is coming soon to CometAPI. Once available, teams will be able to test the model through CometAPI's unified API workflow and compare it directly with existing production models.
While waiting, **Kimi K3 **can serve as a practical alternative for advanced coding, reasoning, and agent workflows. Teams can start building their evaluation baseline with Kimi K3 today, then rerun the same tasks on Qwen3.8 Max when it becomes available.
For teams already using the Qwen family, Qwen3.7 Max API remains a useful reference point for measuring quality, latency, and cost improvements across model generations.
You can also browse the CometAPI Model Catalog for the latest model availability.