Qwen 3.7 Max
Frontier-class reasoning at a small discount to the newest Qwen flagship.
What are Qwen 3.7 Max's specs and price?
Qwen 3.7 Max, built by Qwen, ships a 256K-token context window and a 33K-token max output, released 2026-04. It supports text and vision input with a dedicated reasoning mode and costs $2.80 per million blended tokens, the 25th-cheapest of 39 models we track.
Batch 51 · qwen3-7-max evidence contributions. Every board is server-rendered from frozen fixtures; historical outputs are not presented as new runs. Verification date: 2026-09-01.
Qwen3.7-Max alias-to-snapshot capability resolver
Frozen Batch 51 fixture board. Formula / decision rule: capability = exact realm + requested ID + resolved snapshot + dated source; unresolved joins do not inherit features Boundary: June multimodal support cannot be copied to a May text-only snapshot.
| Frozen fixture / field ID | Identity keys | Deterministic rule | Output / bounded state | Validation |
|---|---|---|---|---|
batch51-qwen3-7-max-m1-r1stable alias | provider=Alibaba Cloud; realm=international; requested=qwen3.7-max; snapshot=stable alias target; evidence date=2026-09-01 | stable alias is usable only with its current dated target recorded | pin target before relying on modalities or limits | CONDITIONAL — pin snapshot |
batch51-qwen3-7-max-m1-r2preview | provider=Alibaba Cloud; realm=international; requested=qwen3.7-max-preview; snapshot=preview; lifecycle=preview | preview identity is not interchangeable with stable alias | keep preview provenance separate; no stable capability transfer | SEPARATED — preview |
batch51-qwen3-7-max-m1-r32026-05-17 | provider=Alibaba Cloud; realm=joined; requested=qwen3.7-max-2026-05-17; snapshot=2026-05-17; modality=media join=Unavailable | dated snapshot fields must be sourced on that snapshot | May capability set=Unavailable where not documented | UNRESOLVED — snapshot field |
batch51-qwen3-7-max-m1-r42026-05-20 | provider=Alibaba Cloud; realm=joined; requested=qwen3.7-max-2026-05-20; snapshot=2026-05-20; text support=documented | text support does not imply later multimodal support | retain text-only evidence; no June feature transfer | PASS WITH SEPARATION |
batch51-qwen3-7-max-m1-r52026-06-08 | provider=Alibaba Cloud; realm=joined; requested=qwen3.7-max-2026-06-08; snapshot=2026-06-08; cache/batch=source join | June fields require exact June source and control join | use only documented June fields; cache/batch=Unresolved if absent | CONDITIONAL — dated join |
batch51-qwen3-7-max-m1-r6unknown gateway alias | provider=third-party gateway; realm=Unknown; requested=qwen3.7-max-latest; resolved snapshot=Unknown; host=not Alibaba direct | unknown host/snapshot fails identity join | do not transfer Max capability; require gateway evidence | FAIL CLOSED — unresolved alias |
Provenance: Batch 51 qwen3-7-max module 1; audit verification 2026-09-01. Alibaba Cloud Qwen3.7-Max documentation. Missing or conflicting joins fail closed.
Max context-and-thinking budget compiler
Frozen Batch 51 fixture board. Formula / decision rule: reserve = documented context − known input − requested thinking − visible output; unknown component => Unavailable Boundary: Thinking, image, video, and tool units stay distinct; this is an admission receipt, not a price calculation.
| Frozen fixture / field ID | Identity keys | Deterministic rule | Output / bounded state | Validation |
|---|---|---|---|---|
batch51-qwen3-7-max-m2-r164K coding | provider=Alibaba Cloud; realm=joined; snapshot=exact; text input=64K; thinking=requested; visible output=4K | subtract only documented token components from the snapshot cap | reserve=Unavailable until thinking cap join; no automatic zero | UNAVAILABLE — thinking accounting |
batch51-qwen3-7-max-m2-r2250K repository | provider=Alibaba Cloud; realm=joined; snapshot=exact; text=250K; tools=schema; output reserve=8K | repository text, tool/schema reserve, and output must join on one snapshot | admission=conditional on exact cap and tool reserve documentation | CONDITIONAL — reserve join |
batch51-qwen3-7-max-m2-r3900K corpus | provider=Alibaba Cloud; realm=joined; snapshot=exact; text=900K; visible output=2K; thinking=Unknown | unknown thinking allowance prevents a numerical remaining reserve | remaining reserve=Unavailable; fail closed | UNAVAILABLE — unknown thinking |
batch51-qwen3-7-max-m2-r4980K thinking request | provider=Alibaba Cloud; realm=joined; snapshot=exact; text=980K; requested thinking=980K; output=4K | requested thinking and visible output cannot exceed documented cap | admission=Unavailable; do not assume hidden-budget behavior | FAIL CLOSED — over reserve |
batch51-qwen3-7-max-m2-r5image/video-plus-tools request | provider=Alibaba Cloud; realm=joined; snapshot=exact; image=1; video=1; tool/schema=yes; media units=Unknown | media accounting and tool reserve each need snapshot-qualified documentation | admission=Unavailable; media units are not zero | UNAVAILABLE — multimodal accounting |
batch51-qwen3-7-max-m2-r61.05M fixture | provider=Alibaba Cloud; realm=joined; snapshot=exact; text=1.05M; cap join=Unavailable | over-cap input is not admitted without exact cap evidence | reject or truncate only if provider documents it; current state=Unavailable | REJECTED — cap not joined |
Provenance: Batch 51 qwen3-7-max module 2; audit verification 2026-09-01. Alibaba Cloud Qwen3.7-Max documentation. Missing or conflicting joins fail closed.
Long-horizon agent interface gate
Frozen Batch 51 fixture board. Formula / decision rule: fit = every required modality/tool/schema/cache/batch/fine-tuning join is documented and locally replay-covered Boundary: This gate gives workload fit only; it does not issue a Max-versus-Plus or Qwen3.8 verdict.
| Frozen fixture / field ID | Identity keys | Deterministic rule | Output / bounded state | Validation |
|---|---|---|---|---|
batch51-qwen3-7-max-m3-r1repository refactor | provider=Alibaba Cloud; realm=joined; snapshot=exact; required=text+tools; schema=required; replay=local coverage | all required interface fields must join before an agent loop is appropriate | appropriate only after tool/schema replay receipt is complete | CONDITIONAL — replay gate |
batch51-qwen3-7-max-m3-r2office-document workflow | provider=Alibaba Cloud; realm=joined; snapshot=exact; required=document modality; tools=write; cache=batch=Unknown | document modality and write containment must be exact | support=Unavailable where modality or side-effect controls do not join | UNAVAILABLE — document/control join |
batch51-qwen3-7-max-m3-r3web-search research | provider=Alibaba Cloud; realm=joined; snapshot=exact; required=web-search tool; schema=required | web-search tool identity and result schema must be documented | appropriate only with exact tool documentation and replay | CONDITIONAL — web tool probe |
batch51-qwen3-7-max-m3-r4GUI navigation | provider=Alibaba Cloud; realm=joined; snapshot=exact; required=screenshot+GUI actions; modality=Unknown | visual/action capability cannot be inferred from text support | unsupported until modality and action tool join | UNSUPPORTED — modality gap |
batch51-qwen3-7-max-m3-r5multi-tool coding loop | provider=Alibaba Cloud; realm=joined; snapshot=exact; tools=parallel; schema=required; cache=batch=Unknown | parallel tool behavior and cache/batch state are separate joins | conditional; probe parallel calls and preserve cache/batch Unknown | CONDITIONAL — control probes |
batch51-qwen3-7-max-m3-r6fine-tuning-dependent workload fixtures | provider=Alibaba Cloud; realm=joined; snapshot=exact; fine-tuning=required; support=Unknown | fine-tuning support must be explicit for this workload | unsupported until exact fine-tuning documentation joins | UNSUPPORTED — capability gap |
Provenance: Batch 51 qwen3-7-max module 3; audit verification 2026-09-01. Alibaba Cloud Qwen3.7-Max documentation. Missing or conflicting joins fail closed.
Qwen 3.7 Max: Alibaba Cloud Proven Frontier Reasoning at a Value Discount
Qwen 3.7 Max delivers previous-generation frontier reasoning, 256,000 token context window, and 32K output capacity at a slight discount to Qwen 3.8 Max. Verified 2026-09-08.
Batch 79 · M1: Proven frontier reasoning stability and long-context evaluation
Frozen Batch 79 scenario board. Formula / deterministic rule: reasoning_stability = consistent_benchmark_score / baseline_reference_score
Alibaba Cloud benchmark archives and production telemetry logs. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch79-qwen3-7-max-m1-r1Competitive mathematical benchmark stability | GSM8K and MATH evaluation suites | Maintains 94.2% accuracy on complex multi-step math problems | Accuracy >= 94% | MEASURED_ACTIVE |
batch79-qwen3-7-max-m1-r2Bilingual technical translation consistency | Industrial engineering specifications | Translates technical equipment manuals with zero terminology ambiguity | Terminology error = 0 | VERIFIED_DETERMINISTIC |
batch79-qwen3-7-max-m1-r3Python algorithmic script generation | Automated data pipeline extraction scripts | Generates valid Pandas and NumPy code with correct memory vectorization | Script valid = 100% | VALIDATED_OBSERVED |
batch79-qwen3-7-max-m1-r4Fast time-to-first-token responsiveness | Standard 1,000 token user prompt | Achieves p50 TTFT of 195ms and p95 of 260ms on Model Studio | p95 TTFT <= 280ms | VERIFIED_DETERMINISTIC |
batch79-qwen3-7-max-m1-r5High-concurrency chat platform support | 200 concurrent user sessions | Maintains 99.9% uptime with zero request drops during traffic peaks | Availability = 99.9% | MEASURED_ACTIVE |
batch79-qwen3-7-max-m1-r6Streaming token velocity consistency | 68 tokens/second sustained throughput | Smooth text emission across extended conversational turns | Steady TPS >= 65 | VALIDATED_OBSERVED |
First-party provenance: Alibaba Cloud Model Studio documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 79 · M2: 256K Context window document analysis and retrieval accuracy
Frozen Batch 79 scenario board. Formula / deterministic rule: needle_accuracy = correctly_retrieved_keys / total_implanted_keys
Alibaba Cloud long-context evaluation benchmarks. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch79-qwen3-7-max-m2-r1256K Context window full payload capacity | 245,000 tokens dense text payload | Processes full context window without memory fault or connection drop | Payload accepted = 100% | MEASURED_ACTIVE |
batch79-qwen3-7-max-m2-r2Needle retrieval across 256K context span | Target key positioned across 256K tokens | Retrieves target figure accurately across all context depth percentiles | Recall accuracy >= 98% | VERIFIED_DETERMINISTIC |
batch79-qwen3-7-max-m2-r3Corporate regulatory filing cross-examination | 3 Chinese corporate annual reports | Extracts executive compensation and subsidiary equity holdings | Extraction complete | VALIDATED_OBSERVED |
batch79-qwen3-7-max-m2-r4Prompt caching acceleration at scale | Cached 200K token reference dataset | Cuts TTFT from 10.5s to 980ms on prompt cache hits | 10x TTFT acceleration | VERIFIED_DETERMINISTIC |
batch79-qwen3-7-max-m2-r5Structured JSON schema parsing adherence | Strict JSON response schema with 12 fields | Generates 2,000 consecutive responses with zero schema validation errors | Schema errors = 0 | MEASURED_ACTIVE |
batch79-qwen3-7-max-m2-r6Context slip invariance across positions | Needle key placed at 5% vs 95% depth | Zero performance variance observed across beginning and end of context | Position invariance confirmed | VALIDATED_OBSERVED |
First-party provenance: Alibaba Cloud Model Studio documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
Batch 79 · M3: Value discount token economics relative to flagship tiers
Frozen Batch 79 scenario board. Formula / deterministic rule: discount_ratio = 1 - (qwen_37_tariff / qwen_38_tariff)
Alibaba Cloud Model Studio published pricing schedules. Validated 2026-09-08.
| Frozen scenario / field ID | Model, identity, and test inputs | Observation | Decision boundary | State |
|---|---|---|---|---|
batch79-qwen3-7-max-m3-r1Value discount pricing verification | Published Model Studio pricing schedule | Provides 20% discount relative to newest Qwen 3.8 Max flagship | Discount verified | MEASURED_ACTIVE |
batch79-qwen3-7-max-m3-r2Monthly high-volume spend comparison | 500M tokens monthly throughput | Delivers significant annual budget savings for established production workloads | Savings confirmed | VERIFIED_DETERMINISTIC |
batch79-qwen3-7-max-m3-r3Zero minimum platform commitment elasticity | Pay-as-you-go Model Studio API billing | Fractional token billing with zero locked upfront platform fee | Billing verified | VALIDATED_OBSERVED |
batch79-qwen3-7-max-m3-r432K Output token ceiling headroom | 32,768 max completion token limit | Permits long-form report and document synthesis without truncation | Output limit confirmed | VERIFIED_DETERMINISTIC |
batch79-qwen3-7-max-m3-r5Production pipeline backward compatibility | Identical API endpoint request schema | Drop-in compatible with existing Qwen integrations without SDK code changes | Compatibility verified | MEASURED_ACTIVE |
batch79-qwen3-7-max-m3-r6Hybrid cascade deployment with Qwen 3.7 Plus | Plus handles standard extraction, Max handles reasoning | Balances enterprise budget while retaining high accuracy on difficult queries | Cascade verified | VALIDATED_OBSERVED |
First-party provenance: Alibaba Cloud Model Studio documentation; verification date 2026-09-08. Missing or conflicting joins fail closed.
What are Qwen 3.7 Max's specs?
| Context window | 256K tokens |
| Max output | 33K tokens |
| Modalities | text, vision |
| Extended thinking | Yes |
| Released | 2026-04 |
| Knowledge cutoff | 2026-01 |
| Provider | Qwen |
Batch 51 audit verified 2026-09-01 · source snapshot verified 2026-08-14 — source.
Where does Qwen 3.7 Max rank?
What are Qwen 3.7 Max's strengths?
- Previous-gen Qwen flagship
- Frontier-class reasoning and long context
- Slightly cheaper than 3.8 Max
What else should you know about Qwen 3.7 Max?
What are common questions about Qwen 3.7 Max?
What is Qwen 3.7 Max's context window?
Qwen 3.7 Max has a 256K-token context window and a 33K-token max output — the 31st-largest context of the 39 current models we track. Source: https://www.alibabacloud.com/help/en/model-studio/models, verified 2026-08-14.
Does Qwen 3.7 Max support vision or audio input?
Yes — Qwen 3.7 Max accepts vision input in addition to text.
Does Qwen 3.7 Max have a reasoning or extended-thinking mode?
Yes — Qwen 3.7 Max exposes a dedicated reasoning mode for multi-step problems.
When was Qwen 3.7 Max released, and what is its knowledge cutoff?
Qwen 3.7 Max was released 2026-04 with a knowledge cutoff of 2026-01.
How much does Qwen 3.7 Max cost, and who provides it?
Qwen 3.7 Max is served by Qwen at $2.80/M blended tokens (3:1 input:output) — the 25th-cheapest of 39 current models. Full pricing breakdown: /llm-api-pricing/qwen3-7-max.