New: Qwen4-Max eval report is here.Read the report ↗

Across the task benchmarks we track, Qwen4-Max’s strongest relative results are in work automations. On AutomationBench-AA, it scores 72.56%, ranking #6 among 58 models with reported results in our dataset.

Updated: Data as of:

Strongest results

Strongest results for Qwen4-Max in tracked evaluation cohorts
BenchmarkMeasuresScore
AA-Briefcase ↗1.1 / combined EloBusiness deliverables1,686.07 Elo#4 of 60
AutomationBench-AA ↗1.0.6 / AA private 657Work automations72.56%#6 of 58
GDP.pdf ↗100 tasks / All-pass / five attemptsProfessional document reasoning26.2%#11 of 58
Terminal-Bench ↗4.0 / mini-swe-agentWorking in a terminal48.48%#14 of 59

Model details

Proprietary22 benchmarks
Specifications for Qwen4-Max
ModelQwen4-Max
Agent Index v1.3Insufficient coverage (4/7) · Preview
CreatorQwen (opens in a new tab)
AvailabilityProprietary
Context size1,000,000 tokens (opens in a new tab)
Max output262,144 tokens (opens in a new tab)
Knowledge cutoffNot reported
LicenseNot reported

Pricing and performance

Per provider and reported configuration. Prices are USD per 1M tokens.

Provider offerings for Qwen4-Max
Provider / configurationContexttokensInputUSD / 1M tokensOutputUSD / 1M tokensLatencyTTFT · secondsSpeedtokens / sec
Standard API, Singapore; uncached input1,000,000$1.50 (opens in a new tab)$4.50 (opens in a new tab)Not reportedNot reported

Blended token cost

USD per 1M tokens · 50% input + 50% output · lower is better

Qwen4-Max highlighted10 of 47 priced models selected
Pricing details and sources

Blended cost = (input price + output price) ÷ 2: 500,000 input and 500,000 output tokens. Each model uses its lowest sourced current provider rate, excluding batch and Flex tiers. Cache discounts, tool fees and workload-specific token usage are excluded. 52 models without complete sourced current prices are unavailable in this comparison. The default selection is this model and the nine closest in blended cost; if this model has no price, we show the ten cheapest. The scale stays fixed across pages of your selection.

Pricing sources for selected models
ModelInput / 1MOutput / 1MBlended / 1MProvider / source
Gemini 3.8 Flash$0.75$3.75$2.25Google AI StudioGemini 3.8 Flash (high) AI Studio · Checked 2026-11-02
Kimi K2.6$0.95$4.00$2.475Moonshot AIKimi K2.6 (Reasoning) · Checked 2026-11-02
Inkling$1.00$4.05$2.525Thinking Machines ↗Serverless inference beta, 256K context · Checked 2026-11-02
Muse Spark 1.1$1.25$4.25$2.75MetaMuse Spark 1.1 (xhigh) · Checked 2026-11-02
Muse Spark 1.2$1.25$4.25$2.75MetaMuse Spark 1.2 (xhigh) · Checked 2026-11-02
Muse Spark 1.3$1.25$4.25$2.75MetaMuse Spark 1.3 (max) · Checked 2026-11-02
GLM-5.2$1.40$4.40$2.90Z.aiGLM-5.2 (Max) · Checked 2026-11-02
GPT-6.1 Terra$1.00$5.00$3.00OpenAI ↗Standard API, prompts up to 272K input tokens · Checked 2026-11-02
Qwen4-Max$1.50$4.50$3.00QwenCloud ↗Standard API, Singapore; uncached input · Checked 2026-11-16
Grok 4.7$2.00$6.00$4.00SpaceXAIGrok 4.7 (xhigh) · Checked 2026-11-02

Output speed

Output tokens per second · higher is better

No sourced output speed for Qwen4-Max.10 of 34 models with reports selected
Measurement details and sources

Output speed measures generated tokens per second. Each model uses its fastest sourced current provider report. Batch and Flex tiers are excluded. Provider, configuration, prompt and reporting methods can differ, so these are reported observations rather than controlled head-to-head tests. 65 models have no qualifying sourced speed report. Defaults include this model and the nine closest reported values, or the ten leading models if this model has no report. The scale stays fixed across selected pages.

Output speed sources for selected models
ModelOutput speedProvider / configurationMeasurementSource
DeepSeek V4.1 Flash 10171,075.275 t/sInco (FAST)DeepSeek V4.1 Flash 1017 (max) (FAST)longmedianArtificial Analysis: DeepSeek V4.1 Flash 1017 (max) (FAST); host Inco (FAST); standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
DeepSeek V4.1 Flash830.034 t/sBaseten (FAST)DeepSeek V4.1 Flash (max) (FAST)longmedianArtificial Analysis: DeepSeek V4.1 Flash (max) (FAST); host Baseten (FAST); standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
MiniMax M3.1 Flash812.597 t/sIncoMiniMax-M3.1-FlashlongmedianArtificial Analysis: MiniMax-M3.1-Flash; host Inco; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
GLM-5.3688.217 t/sSystalyzeGLM-5.3 (max)longmedianArtificial Analysis: GLM-5.3 (max); host Systalyze; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
Kimi K3481.694 t/sInco (FAST)Kimi K3 (max) (FAST)longmedianArtificial Analysis: Kimi K3 (max) (FAST); host Inco (FAST); standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
GPT-6.1 Sol342.946 t/sOpenAI (Ultrafast)GPT-6.1 Sol (max) (Ultrafast)longmedianArtificial Analysis: GPT-6.1 Sol (max) (Ultrafast); host OpenAI (Ultrafast); standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
DeepSeek V4 Pro 0813305.002 t/sSystalyzeDeepSeek V4 Pro 0813 (max)longmedianArtificial Analysis: DeepSeek V4 Pro 0813 (max); host Systalyze; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
Muse Spark 1.1294.835 t/sMetaMuse Spark 1.1 (xhigh)longmedianArtificial Analysis: Muse Spark 1.1 (xhigh); host Meta; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
Muse Spark 1.2277.574 t/sMetaMuse Spark 1.2 (xhigh)longmedianArtificial Analysis: Muse Spark 1.2 (xhigh); host Meta; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
Gemini 4 Hydrogen198.156 t/sGoogleStandard Gemini API; independent speed measured at xhighPrompt not specifiedStatistic not specifiedOfficial standard API price; output includes thinking. Explicit caching unsupported. Independent first-party Google output-speed headline at xhigh. 41.11s is time to first answer, so it is not entered as TTFT. No region/concurrency/measurement date was reported in this summary.SourceChecked 2026-11-12

Latency

Time to first token in seconds · lower is better

No sourced time-to-first-token latency for Qwen4-Max.10 of 32 models with reports selected
Measurement details and sources

Latency measures time to first token (TTFT), converted to seconds. Each model uses its lowest sourced current TTFT report; total response time is excluded. Batch and Flex tiers are excluded. Provider, configuration, prompt and reporting methods can differ, so these are reported observations rather than controlled head-to-head tests. 67 models have no qualifying sourced latency report. Defaults include this model and the nine closest reported values, or the ten leading models if this model has no report. The scale stays fixed across selected pages.

Latency sources for selected models
ModelLatencyProvider / configurationMeasurementSource
DeepSeek V4.1 Flash 10170.182 sBaseten (FAST)DeepSeek V4.1 Flash 1017 (max) (FAST)longmedianArtificial Analysis: DeepSeek V4.1 Flash 1017 (max) (FAST); host Baseten (FAST); standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
DeepSeek V4.1 Flash0.311 sTogether AIDeepSeek V4.1 Flash (max)longmedianArtificial Analysis: DeepSeek V4.1 Flash (max); host Together AI; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
MiniMax M3.1 Flash0.584 sTogether AIMiniMax-M3.1-FlashlongmedianArtificial Analysis: MiniMax-M3.1-Flash; host Together AI; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
GLM-5.30.597 sTogether AIGLM-5.3 (max)longmedianArtificial Analysis: GLM-5.3 (max); host Together AI; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
MiniMax M30.699 sMiniMaxMiniMax-M3source aggregatemedianExplicit first-party AA measurement; reasoning configuration retained. Cross-provider medians are excluded. Retrieval time is not measurement time.SourceChecked 2026-11-02
DeepSeek V4 Pro 08130.832 sDigitalOceanDeepSeek V4 Pro 0813 (max)longmedianArtificial Analysis: DeepSeek V4 Pro 0813 (max); host DigitalOcean; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
Kimi K30.87 sDatabricksKimi K3 (max)longmedianArtificial Analysis: Kimi K3 (max); host Databricks; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
Claude Opus 4.61.35 sAmazonClaude Opus 4.6 (non-reasoning, high)longmedianArtificial Analysis: Claude Opus 4.6 (non-reasoning, high); host Amazon; standard uncached input/output pricing. TTFT is the first streamed token, which may be reasoning. Retrieval time is not measurement time; region/concurrency/quantization not specified in this record.SourceChecked 2026-11-02
GLM-5.3-Flash1.546 sZ.aiGLM-5.3-Flashsource aggregatemedianExplicit first-party AA measurement; reasoning configuration retained. Cross-provider medians are excluded. Retrieval time is not measurement time.SourceChecked 2026-11-02
GLM-5.21.777 sZ.aiGLM-5.2 (Max)source aggregatemedianExplicit first-party AA measurement; reasoning configuration retained. Cross-provider medians are excluded. Retrieval time is not measurement time.SourceChecked 2026-11-02

Benchmark performance

Qwen4-Max in color, alongside leading models. Latest stored reports, best effort per report date.

Reported benchmarks

All published results in our dataset, with versions, evaluation settings, and sources.

All published benchmark reports for Qwen4-Max
BenchmarkScoreEvaluation setupSource
AA-Briefcase1.1 / combined EloCombined Elo · Higher is better1,686.07 Eloeffort: xhigh · Exact independent xhigh record. Combined Elo, not rubric pass percentage. · Tools enabledArtificial Analysis (opens in a new tab)independent
AgentArenaQwen4 launch snapshotNet improvement · Higher is better7.5 pointsLaunch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
Agents’ Last Exam1.1 / Qwen4 launch / task release unspecifiedReported score · Higher is better46.7%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
APEX-Agents1.1 / Qwen4 launch snapshotReported score · Higher is better72.1%Launch comparison; effort and full harness unspecified. Launch cites Mercor leaderboard; exact snapshot retained.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
AutomationBench-AA1.0.6 / AA private 657Objectives completed without guardrail violations · Higher is better72.561%effort: xhigh · Exact independent xhigh record; protocol per archived Artificial Analysis methodology. Effort: xhigh. · Tools enabledArtificial Analysis (opens in a new tab)independent
BioMysteryBenchQwen4 launch / protocol unspecifiedReported score · Higher is better85%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
ChartographyQwen4 launch / min_pixels 4096×32×32Reported score · Higher is better54.2%Launch comparison; effort and full harness unspecified. min_pixels=4096×32×32; tool access not specified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
CodeArenaQwen4 launch snapshotElo · Higher is better1,750 EloLaunch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
DeepSWEv1.1 / Qwen4 launchPass@1 · Higher is better75.7%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
FrontierCode2.0 / independent mini-swe-agentAccuracy · Higher is better26.453%effort: xhigh · Exact independent xhigh record; protocol per archived Artificial Analysis methodology. Effort: xhigh. · Tools enabledArtificial Analysis (opens in a new tab)independent
GDP.pdf100 tasks / All-pass / five attemptsAccuracy · Higher is better26.2%effort: xhigh · Exact independent xhigh record; protocol per archived Artificial Analysis methodology. Effort: xhigh. · No toolsArtificial Analysis (opens in a new tab)independent
HealthBench ProfessionalQwen4 launch / protocol unspecifiedReported score · Higher is better60%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
HLAB-Vals1.1 / Qwen4 launch snapshotReported score · Higher is better21.3%Launch comparison; effort and full harness unspecified. Launch cites Vals Index; do not equate with Harvey LAB-AA.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
HLE-Diamond1,000 questions / no toolsAccuracy · Higher is better44.9%effort: xhigh · Exact independent xhigh record; protocol per archived Artificial Analysis methodology. Effort: xhigh. · No toolsArtificial Analysis (opens in a new tab)independent
Humanity's Last ExamHLE — source does not specify subsetAccuracy · Higher is better51.7%Qwen launch chart labels Humanity’s Last Exam; subset, tool access and effort unspecified. Not HLE-Diamond.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
JobBenchQwen4 launch / updated dataset and evaluatorReported score · Higher is better68.5%Launch comparison; effort and full harness unspecified. Dataset commit d43a66575635da0d3f6cb9a20976dc1a60f3ffc9; evaluator 660b33a86ed032559df0b2fdc1ad38cf7ceb83f3.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
ProgramBenchQwen4 launch / protocol unspecifiedReported score · Higher is better31.5%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
Terminal-Bench4.0 / Qwen4 launch / harness unspecifiedTask success · Higher is better48.5%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
Terminal-Bench4.0 / mini-swe-agentTask success · Higher is better48.485%effort: xhigh · Exact independent xhigh record; protocol per archived Artificial Analysis methodology. Effort: xhigh. · Tools enabledArtificial Analysis (opens in a new tab)independent
Terminal-Bench-Science0.1 / independent mini-swe-agentAccuracy · Higher is better23.333%effort: xhigh · Exact independent xhigh record; protocol per archived Artificial Analysis methodology. Effort: xhigh. · Tools enabledArtificial Analysis (opens in a new tab)independent
Video-IndexQwen4 launch / protocol unspecifiedReported score · Higher is better68.1%Launch comparison; effort and full harness unspecified.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15
Work360Qwen4 launch / internal evaluationReported score · Higher is better52.69 scoreInternal Work360 evaluation; 26 Tier-1 domains, 182 Tier-2 subdomains and 721 Tier-3 fields; full reproducibility protocol not published in launch text.Qwen Team ↗ (opens in a new tab)provider reported · 2026-11-15