AI Model Reviews

Qwen3.8 Max Tested: One Brilliant Build, Several Expensive Misses

Direct Answer

Qwen3.8 Max looks like a capable frontier model with a high ceiling, but this test does not establish it as the best open model. Bijan Bowen's strongest result was genuinely impressive: a detailed C++ skate simulation set on an early-2000s New York block. The wedding website also combined attractive design with coherent long-form writing. But the model missed core CAD constraints, struggled with reference-image fidelity, produced uneven game logic, and needed repair passes to rescue several builds.

The word open also needs a date attached. Qwen's 3 August announcement says the 2.4T Max model and a 27B companion will receive open weights next week. At publication time, the hosted model is available, while downloadable weights, the final model card, file sizes, and license still need to appear. A promise to publish open weights is meaningful. It is not the same as weights readers can download today.

JQ AI SYSTEMS verdict: test Qwen3.8 Max for difficult C++ prototypes, long-form product experiences, and parallel candidate generation. Keep deterministic verification around geometry, simulation, state logic, and visual fidelity. Do not migrate production work on the strength of launch benchmarks or one excellent demo.

Watch the Full Test

Video credit: Bijan Bowen. Watch the original video on YouTube. Test outcomes and the reported $32 spend are creator observations. Availability was checked against Alibaba Cloud documentation and Qwen's official 3 August announcement.

What Is Actually Released?

ItemStatus on 3 Aug 2026Practical meaning
qwen3.8-maxListed in Alibaba Cloud's current model indexHosted access is available through regional Model Studio endpoints.
qwen3.8-max-previewStill documented on Token Plan and integration pagesOlder subscriptions and examples may retain the preview ID. Check the console for the correct regional endpoint.
Qwen3.8 Max weightsAnnounced for the following weekNo local deployment claim should be treated as complete until files and a license are published.
Qwen3.8 27B weightsAnnounced for the following weekThis is likely the more practical local release, but capability and memory claims remain provisional.
Open-source statusNot establishedOpen weights describe downloadable parameters. Full open source would also require broader training and development artifacts.

Alibaba Cloud's current model index lists qwen3.8-max for OpenAI-compatible, Anthropic-compatible, and DashScope access in Beijing, Singapore, Tokyo, Frankfurt, and Virginia. Its July Token Plan launch page still describes qwen3.8-max-preview as the latest flagship. That documentation lag is a good reason to copy the model ID from your own console instead of a third-party setup guide.

Specs, Benchmarks, and What Is Still a Claim

Qwen's launch material, as shown in the video and linked from its official announcement, describes a 2.4-trillion-parameter Mixture-of-Experts model with about 95 billion parameters active per token. It also presents a one-million-token context window, up to 128K output, image and video understanding, and benchmark results aimed at frontier coding and knowledge-work models.

Those numbers are useful launch claims, but they are not interchangeable with independent evidence. A parameter count does not measure usability. A benchmark score does not tell you whether a printable motor cavity is correct. A million-token context does not guarantee that an agent will preserve the right constraints across a 50-minute build. The video is valuable precisely because it reveals behavior the headline table cannot.

Launch claimHow to read it
2.4T total parametersModel scale and storage burden, not proof of better outcomes.
About 95B active per tokenMoE compute path; it does not mean only 95B weights must be stored.
1M contextLarge input capacity; retrieval quality and long-run instruction retention still need task-specific tests.
128K maximum outputA ceiling, not a sensible default budget for an agent task.
56.6 on DeepSWE 1.1A vendor-reported launch result discussed in the video, not a substitute for your repository's acceptance tests.
10+ days of autonomous codingAn endurance demonstration. Reliability depends on the harness, tools, checkpoints, recovery logic, and verifier.

How Bijan Ran the Test

This is a broad capability probe, not a controlled head-to-head benchmark. Bijan used Qwen's web interface for some one-shot tests and an OpenAI-compatible Qwen endpoint inside Codex CLI for longer builds. He cleared prior cache and memory, generally used fresh sessions, and selected a high reasoning setting. Several projects received follow-up repairs; at least one began in the web app and was then moved into Codex.

Those choices make the video excellent for discovering failure modes and weak for declaring an overall winner. The projects had different time budgets, different repair opportunities, and different harnesses. Some prompts are recurring public tests. A model can also look better in a harness that compiles, opens a browser, inspects screenshots, and retries than it does in a one-shot chat window.

The Ten-Test Scorecard

TestWhat workedWhat failedVerdict
Browser OSThemed apps, games, sound, personalization, and a memorable window effectDrawing cursor offset, one-open-only apps, weak GTA-like gameMixed-positive
C++ NYC skate simRich city, cabs, pedestrians, strong atmosphere, repaired ollie and cameraInitial movement, Z-fighting, and camera issues; needed a second passStandout
Printable RB26 shellDetailed engine-like form and readable twin-cam treatmentWrong turbo orientation, no valid motor cavity, ignored support-free print constraintFailure
Street Yeet gameFunny cutscene, generated dialogue, interesting ragdoll effectsEmpty first city, invisible entities, incomplete world and weak gameplayBelow target
Luxury watch siteCinematic camera movement and a presentable product pageBand orientation, backward time, static hands, blurry glassVisually strong, mechanically wrong
Wedding writing siteCoherent story, recurring book motif, full event details, attractive paletteNo major issue surfaced in the walkthroughStrong
Subway FPSCompetent scene after repairBuggy web output, generic result, misplaced subway carMixed
HumorA serviceable kale-and-chicken jokeOne prompt cannot establish durable comedic abilityAcceptable
C++ Chrono CityAttempted the requested multi-era conceptLong run ended in a poor, hard-to-use resultFailure
Image-to-computer buildInteractive model with spinning fan, LEDs, and auto-orbitWrong color, shape, and reference fidelityCool generic object, failed reconstruction

Why the New York Skate Build Matters

The skate simulation is the strongest evidence in the episode because it was not just a polished landing page. The model had to create C++ code, a navigable environment, a player controller, camera behavior, street life, and visual atmosphere. The first pass was broken, but a fresh repair thread recovered the ollie, reduced Z-fighting, improved the camera, and made the scene feel inhabited. The final project exceeded 2,000 lines and was the one result Bijan treated as a clear winner.

That recovery is more informative than a perfect screenshot. It suggests Qwen3.8 Max can inspect a concrete failure, preserve a substantial codebase, and make a meaningful second pass. For production use, though, the success criterion should be automated: the game launches, controls map correctly, collision tests pass, the frame rate stays above a threshold, and a browser or vision check confirms the expected states.

The Failure Patterns Were Consistent

  1. Spatial plausibility over exact geometry. The watch and computer looked impressive until compared with the requested object. The CAD engine looked detailed while failing the motor-fit requirement.
  2. Presentation over mechanics. Several games had a convincing mood but weak controls, incomplete worlds, or broken state.
  3. Constraint loss over long runs. The printable shell ignored core manufacturability instructions even after almost an hour of work.
  4. Repairability without first-pass reliability. Codex improved multiple outputs, but the cost and time of those repair loops belong in the score.
  5. Generic substitution. When reference matching became difficult, the model produced an independently attractive object rather than the specified one.
Routing rule: use Qwen3.8 Max where a strong candidate has value and verification is cheap. Route exact CAD, regulated calculations, reference-faithful product work, and stateful production software to workflows with deterministic checks and specialist review.

What Did the Full Test Cost?

Bijan reports approximately $32 for the experiments shown. He also discovers that he used a regular Model Studio API key rather than the dedicated Token Plan key. Alibaba's Token Plan documentation says those keys are separate and that the subscription key uses the sk-sp- format. The lesson is operational: confirm which balance an agent is charging before starting parallel 30-to-50-minute builds.

The video cites a pay-as-you-go rate of $2 per million input tokens and $6 per million output tokens. Treat that as the price observed or displayed during the test, because regional billing pages and model status can change. Alibaba's international Token Plan currently advertises limited-time tiers at $6, $18, and $68 per month, with rolling five-hour and seven-day credit limits rather than an unlimited bucket.

The useful metric is cost per accepted result. The $32 endpoint bill excludes the creator's review time, the failed CAD run, the long Chrono City run, and the follow-up repairs. A model that costs less per token can still cost more per usable deliverable if the acceptance rate is low.

Local Hardware Reality: 2.4T Is Not a Laptop Model

Mixture-of-Experts makes inference more efficient by activating only part of the network for each token. It does not remove the need to store the experts. At the announced 2.4T scale, raw weight memory is roughly 4.8 TB at 16-bit, 2.4 TB at 8-bit, or 1.2 TB at 4-bit. Runtime buffers, KV cache, routing, and serving overhead come on top.

Announced model16-bit raw weights8-bit raw weights4-bit raw weightsLikely audience
Qwen3.8 Max, 2.4TAbout 4.8 TBAbout 2.4 TBAbout 1.2 TBDistributed inference providers and specialist labs
Qwen3.8 27BAbout 54 GBAbout 27 GBAbout 13.5 GBHigh-memory desktops and consumer local-AI setups after quantization

These are arithmetic estimates, not deployment recommendations. The 27B model is the one local builders should watch, but its real context capacity, quantization quality, speed, multimodal support, and license cannot be assessed before the release files arrive. Check the official Qwen Hugging Face organization and QwenLM GitHub organization, not a renamed community upload.

A Better Qwen3.8 Evaluation Protocol

  1. Freeze one real task. Use a private bug, feature, design repair, or document workflow from your own work.
  2. Keep the harness constant. Same Codex version, tools, permissions, machine, repository commit, and system instructions for every model.
  3. Write acceptance checks first. Include unit tests, browser assertions, image-difference thresholds, mesh rules, or dimensional tolerances.
  4. Start clean sessions. Do not let a failed attempt teach the next model where the hidden problem is.
  5. Equalize repair budgets. Give every model the same number of retries, maximum wall time, and token allowance.
  6. Track accepted-result economics. Record API cost, elapsed time, human repair minutes, verifier cost, regressions, and final pass rate.
  7. Repeat the run. One-shot variability is real. Run at least three trials before calling a model reliable.
  8. Wait for the license. For local or commercial deployment, verify the exact checkpoint, model card, and license after weights are published.

Claim Audit

ClaimAssessmentWhy
Qwen3.8 Max is the best open model.Not establishedThe test is mixed, and the announced weights were not yet downloadable at publication time.
It is a 2.4T MoE with about 95B active.Launch claimPublished in Qwen's announcement material; a full public model card and weights remain the stronger artifact.
Open weights arrive next week.Official dated commitmentQwen's 3 August post names both Max and 27B. Verify again after the promised window.
It can code autonomously for more than 10 days.Vendor demonstrationEndurance is useful, but the harness, verifier, checkpoints, and intervention rate determine reliability.
The model is excellent at C++.Plausible, task-specificThe skate build was exceptional; Chrono City failed. One category contains both outcomes.
It is strong at CAD.Not supported hereThe engine shell missed the defining fit and printability constraints.
The full test cost $32.Creator-reportedA useful observed bill, but not a normalized price-performance benchmark.
Max will run locally because it is MoE.MisleadingActive compute is lower than total size; raw quantized weights remain around terabyte scale.

Video Chapters

TimeChapterWhat to watch for
00:00IntroRelease framing and open-weight promise.
01:18First lookAvailability, model family, and headline positioning.
03:54Technical lookArchitecture, context, benchmarks, and API price.
05:20Browser OSFeature breadth versus interaction bugs.
10:16C++ skate simulationThe strongest build and its repair pass.
15:013D CADVisual detail versus failed engineering constraints.
18:19Street YeetCinematic flair and incomplete gameplay.
22:23Luxury watch websitePresentation quality and geometric errors.
24:00Creative writing websiteThe second clear success.
30:19Subway FPSWeb failure, Codex repair, generic result.
31:57HumorA tiny and low-confidence writing probe.
32:35Chrono City C++A long run that failed to become usable.
34:07Multimodal codingAttractive generic output, weak reference fidelity.
35:14Results overviewThe creator's mixed final ranking.
37:14CostThe $32 bill and wrong-key discovery.
38:24Closing thoughtsWhy the 27B checkpoint may matter more locally.

Bottom Line

Qwen3.8 Max deserves serious evaluation, not a coronation. It produced one of the better C++ prototypes in Bijan Bowen's model series and a convincing writing-led website. It also delivered an unusable CAD result, a failed Chrono City build, weak reference reconstruction, and several visually polished projects with mechanical mistakes.

The most important next event is not another reaction video. It is the promised publication of the Max and 27B weights, model cards, and licenses. Once those land, teams can measure reproducible self-hosted performance, serving cost, quantization loss, and commercial terms. Until then, use the hosted model as a candidate generator with strong verification, and judge it by cost per accepted result.

Sources and Further Reading

Common questions

Is Qwen3.8 Max the best open model?
Not yet as an established fact. Bijan Bowen found one exceptional C++ build and a strong writing result, but CAD, game logic, and reference fidelity were inconsistent. The weights and license were also only promised for the following week when this article was checked, so the model was not yet independently deployable as an open-weight release.
Are Qwen3.8 Max weights available to download?
Not as of 3 August 2026. Qwen announced that Qwen3.8 Max and a Qwen3.8 27B model would receive open weights the following week. Until the files, model card, and license are published, developers should treat that as a dated release commitment rather than current availability.
What is the difference between Qwen3.8 Max and Qwen3.8 Max Preview?
Alibaba Cloud introduced qwen3.8-max-preview through its Token Plan in July. Its current model index now lists qwen3.8-max across several regions. Older integration and subscription pages may still reference the preview ID, so developers should use the model ID and endpoint shown in their own Model Studio region.
How much did Bijan Bowen spend testing Qwen3.8 Max?
The video reports about $32 for the full set of experiments. Bijan also explains that he accidentally used a regular API key instead of the dedicated Token Plan key, so this is an observed pay-as-you-go bill from one mixed test session, not a controlled cost benchmark.
Can Qwen3.8 Max run on a local workstation?
The announced 2.4-trillion-parameter Max model is data-center scale. Raw weight arithmetic is about 4.8 TB at 16-bit, 2.4 TB at 8-bit, or 1.2 TB at 4-bit before runtime overhead. The announced 27B companion is the more realistic local candidate, but its final files, quantizations, context behavior, and license still need to be checked after release.
What was Qwen3.8 Max best at in this test?
Its strongest result was a C++ early-2000s New York skate simulation that recovered well after a repair pass. A wedding website also showed coherent long-form creative writing and a consistent visual motif. The model was less reliable when exact geometry, physical fit, reference matching, game state, or camera behavior mattered.
Can Qwen3.8 Max be used inside Codex?
Bijan used an OpenAI-compatible Qwen endpoint inside the Codex CLI for several tests. Alibaba documents OpenAI-compatible and Anthropic-compatible endpoints, but model IDs and regional availability can change. Use a separate low-limit key, a non-sensitive repository, and acceptance tests before allowing agentic changes.
Share
X LinkedIn Reddit
Build Yours

Want a system
like this one?

Book a free 30-minute call. We map your situation, identify the highest-impact automation, and figure out if we are a fit.

Book Free 30-min Call