Model A

xAI

67.13/100

Supported · Public rank #31

90% interval 58.5–75.8

Grok 4.20 vs Qwen3.5 397B

Updated September 10, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.

Model B

Qwen3.5 397B

Alibaba

54.7/100

Estimated · Public rank #100

90% interval 43.2–66.2

Decision reading

Grok 4.20 has the higher public score estimate, 67.13 versus 54.7, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.

6 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.

Share or export

Share on X LinkedInSocial card CSV JSON

Which one for your work

Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.

Long documents
Prompts that approach the documented context limit
Grok 4.20
Grok 4.20 has the larger documented context window.
Confidence: documented
Chat turn cost
1K fresh input + 500 output tokens
Qwen3.5 397B
Qwen3.5 397B has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates

Show secondary and unsupported calls

Repository review cost
50K fresh input + 3K output tokens
Qwen3.5 397B
Qwen3.5 397B has the lower estimated token cost for this stated workload. Costs use the listed standard API rates.
Confidence: listed-rates
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
Qwen3.5 397B is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Qwen3.5 397B is scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.
Confidence: limited
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Not enough matched evidence
The page does not recommend a cost winner because at least one model cannot fit the stated workload in one request. Qwen3.5 397B does not fit this workload in one request. Grok 4.20 has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate.
Confidence: rate-fallback

What is actually comparable

Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.

17 Grok 4.20 only

6 Shared

32 Qwen3.5 397B only

Shared results: 6
Grok 4.20 only: 17
Qwen3.5 397B only: 32
Like-for-like categories: 1 / 8

3 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.

Category results, on a stated basis

Each row shows the public-lane category score for both models: the BenchAlign lane for agentic, coding, and knowledge, and the weighted public rows elsewhere, with the same rank each model page reports. A row is like-for-like only when both scores rest on Supported evidence or the same weighted set. Directional and not-comparable rows remain visible, but they never receive a winner.

Multimodal

Like-for-like

Grok 4.20: 34.6; #43/48
Qwen3.5 397B: 63.0; #27/48
Basis: Provisional lane · 2 vs 2 weighted rows
Reading: Qwen3.5 397B leads

Agentic

Directional only

Grok 4.20: 26.7; Supported · #145/152
Qwen3.5 397B: 46.8; Estimated · #71/152
Basis: BenchAlign lane · 4 vs 13 public rows
Reading: Directional only

Coding

Directional only

Grok 4.20: 28.2; Supported · #141/151
Qwen3.5 397B: 50.2; Estimated · #59/151
Basis: BenchAlign lane · 6 vs 3 public rows
Reading: Directional only

Knowledge

Directional only

Grok 4.20: 49.4; Supported · #85/183
Qwen3.5 397B: 50.2; Estimated · #77/183
Basis: BenchAlign lane · 6 vs 6 public rows
Reading: Directional only

Reasoning

Not comparable

Grok 4.20: 34.2; Unranked · 2 rankable rows
Qwen3.5 397B: 59.8; Unranked · 2 rankable rows
Basis: Provisional lane · 2 vs 1 weighted rows
Reading: Not comparable

Math

Not comparable

Grok 4.20: Not ranked
Qwen3.5 397B: 74.2; Unranked · 5 rankable rows
Basis: Provisional lane · 0 vs 2 weighted rows
Reading: Not comparable

Multilingual

Not comparable

Grok 4.20: Not ranked
Qwen3.5 397B: 69.7; #5/12
Basis: Provisional lane · 0 vs 1 weighted rows
Reading: Not comparable

Instruction following

Not comparable

Grok 4.20: Not ranked
Qwen3.5 397B: 0.0; #123/123
Basis: Provisional lane · 0 vs 0 weighted rows
Reading: Not comparable

Category scores with the server-provided evidence basis for Grok 4.20 and Qwen3.5 397B
Category	Grok 4.20	Qwen3.5 397B	Basis	Reading
Multimodal	34.6#43/48	63.0#27/48	Like-for-likeProvisional lane · 2 vs 2 weighted rows	Qwen3.5 397B leads
Agentic	26.7Supported · #145/152	46.8Estimated · #71/152	Directional onlyBenchAlign lane · 4 vs 13 public rows	Directional only
Coding	28.2Supported · #141/151	50.2Estimated · #59/151	Directional onlyBenchAlign lane · 6 vs 3 public rows	Directional only
Knowledge	49.4Supported · #85/183	50.2Estimated · #77/183	Directional onlyBenchAlign lane · 6 vs 6 public rows	Directional only
Reasoning	34.2Unranked · 2 rankable rows	59.8Unranked · 2 rankable rows	Not comparableProvisional lane · 2 vs 1 weighted rows	Not comparable
Math	Not ranked	74.2Unranked · 5 rankable rows	Not comparableProvisional lane · 0 vs 2 weighted rows	Not comparable
Multilingual	Not ranked	69.7#5/12	Not comparableProvisional lane · 0 vs 1 weighted rows	Not comparable
Instruction following	Not ranked	0.0#123/123	Not comparableProvisional lane · 0 vs 0 weighted rows	Not comparable

Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign) differ from the provisional-lane categories. Unranked scores sit on the lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.

Shape of the matched evidence

Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.

Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.

CharXiv
Multimodal
Grok 4.20: 60.9%Qwen3.5 397B: 80.8%Normalized gap 19.9Grok 4.20 source Qwen3.5 397B source
Terminal-Bench 2.0
Agentic
Grok 4.20: 47.1%Qwen3.5 397B: 52.5%Normalized gap 5.4Grok 4.20 source Qwen3.5 397B source
MMMU-Pro
Multimodal
Grok 4.20: 75.2%Qwen3.5 397B: 79%Normalized gap 3.8Grok 4.20 source Qwen3.5 397B source
SWE-bench Pro
Coding
Grok 4.20: 51.8%Qwen3.5 397B: 50.9%Normalized gap 0.9Grok 4.20 source Qwen3.5 397B source
SWE-bench Verified
Coding
Grok 4.20: 76.7%Qwen3.5 397B: 76.2%Normalized gap 0.5Grok 4.20 source Qwen3.5 397B source

What each workload costs

Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.

Chat turn

1K fresh input + 500 output tokens

Grok 4.20: $0.005; Fits in one request
Qwen3.5 397B: $0.0024; Fits in one request

Qwen3.5 397B has the lower modeled cost

Costs use the listed standard API rates.

Repository review

50K fresh input + 3K output tokens

Grok 4.20: $0.118; Fits in one request
Qwen3.5 397B: $0.0408; Fits in one request

Qwen3.5 397B has the lower modeled cost

Costs use the listed standard API rates.

Cache-heavy agent loop

200K cached + 20K fresh input + 10K output tokens

Grok 4.20: $0.5; Fits in one request; Cached input priced at the published list-input rate
Qwen3.5 397B: $0.168; Does not fit in one request; Cached input priced at the published list-input rate

Qwen3.5 397B does not fit this workload in one request. Grok 4.20 has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate.

Specification differences

Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.

SpecificationGrok 4.20Qwen3.5 397B

Context window

Maximum documented context; output-token limits may be lower.

Grok 4.20

Qwen3.5 397B

128K

API model ID

Grok 4.20

Not sourced

Qwen3.5 397B

Not sourced

Cached-input rate

A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.

Grok 4.20

Not published

Qwen3.5 397B

Not published

Documented inputs

Grok 4.20

Not sourced

Qwen3.5 397B

Not sourced

Documented outputs

Grok 4.20

Not sourced

Qwen3.5 397B

Not sourced

Provider availability

Grok 4.20

Not sourced

Qwen3.5 397B

Not sourced

Reasoning profile

Grok 4.20

Reasoning

Qwen3.5 397B

Non-Reasoning

Weight access

Grok 4.20

Proprietary

Qwen3.5 397B

Open Weight

License

Grok 4.20

Proprietary

Qwen3.5 397B

Open Weight

Release date

Grok 4.20

2026-03-10

Qwen3.5 397B

2026-02-16

If you already use one of these models

Deployment change: The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
Quality signal: Grok 4.20 has the higher public score estimate, 67.13 versus 54.7, but the 90% score intervals overlap.
Workload cost: Repository review: $0.118 vs $0.0408. Cache-heavy agent loop: $0.5 vs $0.168.
Context tradeoff: Grok 4.20 has the larger documented window (2M).

Run the same representative tasks against both endpoints before changing production traffic.

Benchmark evidence

The full public result ledger is available for audit without forcing a wide desktop table onto a phone.

Browse raw public benchmark evidence55 rows

Agentic

Terminal-Bench 2.0
Grok 4.2047.1%
Source
Qwen3.5 397B52.5%
Source
Qwen3.5 397B leads this result
DeepSearchQA
Grok 4.2062.8%
Source
Qwen3.5 397B—
Not directly comparable
Gert Labs
Shared source
Grok 4.2038.36%
Qwen3.5 397B46.76%
Qwen3.5 397B leads this result
Terminal-Bench 2.1 (Vals)
Grok 4.2044.2%
Source
Qwen3.5 397B—
Not directly comparable
BrowseComp
Grok 4.20—
Qwen3.5 397B62%
Source
Not directly comparable
Claw-Eval
Grok 4.20—
Qwen3.5 397B56.8%
Source
Not directly comparable
QwenClawBench
Grok 4.20—
Qwen3.5 397B51.8%
Source
Not directly comparable
τ³-bench results
Grok 4.20—
Qwen3.5 397B68.4%
Source
Not directly comparable
VITA-Bench
Grok 4.20—
Qwen3.5 397B43.7%
Source
Not directly comparable
DeepPlanning
Grok 4.20—
Qwen3.5 397B37.6%
Source
Not directly comparable
Toolathlon
Grok 4.20—
Qwen3.5 397B36.3%
Source
Not directly comparable
MCP Atlas
Grok 4.20—
Qwen3.5 397B46.1%
Source
Not directly comparable
MCP-Tasks
Grok 4.20—
Qwen3.5 397B74.2%
Source
Not directly comparable
WideResearch
Grok 4.20—
Qwen3.5 397B74.0%
Source
Not directly comparable
ResearchClawBench
Grok 4.20—
Qwen3.5 397B14.2%
Source
Not directly comparable

Coding

LiveCodeBench Pro
Grok 4.2074.2%
Source
Qwen3.5 397B—
Not directly comparable
SWE-bench Verified
Grok 4.2076.7%
Source
Qwen3.5 397B76.2%
Source
Grok 4.20 leads this result
SWE-bench Pro
Grok 4.2051.8%
Source
Qwen3.5 397B50.9%
Source
Grok 4.20 leads this result
Vibe Code Bench
Grok 4.204.06%
Source
Qwen3.5 397B—
Not directly comparable
LiveCodeBench (Vals)
Grok 4.2084.3%
Source
Qwen3.5 397B—
Not directly comparable
SWE-bench (Vals)
Grok 4.2072.2%
Source
Qwen3.5 397B—
Not directly comparable
LiveCodeBench v6
Grok 4.20—
Qwen3.5 397B83.6%
Source
Not directly comparable

Reasoning

ARC-AGI-2
Grok 4.2053.3%
Source
Qwen3.5 397B—
Not directly comparable
ARC-AGI-3
Grok 4.200.1%
Source
Qwen3.5 397B—
Not directly comparable
LongBench v2
Grok 4.20—
Qwen3.5 397B63.2%
Source
Not directly comparable
AI-Needle
Grok 4.20—
Qwen3.5 397B68.7%
Source
Not directly comparable

Knowledge

GPQA-D
Grok 4.2088.5%
Source
Qwen3.5 397B—
Not directly comparable
HLE w/o tools
Grok 4.2031.6%
Source
Qwen3.5 397B—
Not directly comparable
HealthBench Hard
Grok 4.2020.3%
Source
Qwen3.5 397B—
Not directly comparable
MedXpertQA (Text)
Grok 4.2050.2%
Source
Qwen3.5 397B—
Not directly comparable
GPQA Diamond (Vals)
Grok 4.2088.6%
Source
Qwen3.5 397B—
Not directly comparable
MMLU-Pro (Vals)
Grok 4.2086.3%
Source
Qwen3.5 397B—
Not directly comparable
GPQA
Grok 4.20—
Qwen3.5 397B88.4%
Source
Not directly comparable
SuperGPQA
Grok 4.20—
Qwen3.5 397B70.4%
Source
Not directly comparable
MMLU-Pro
Grok 4.20—
Qwen3.5 397B87.8%
Source
Not directly comparable
MMLU-Redux
Grok 4.20—
Qwen3.5 397B94.9%
Source
Not directly comparable
C-Eval
Grok 4.20—
Qwen3.5 397B93%
Source
Not directly comparable
HLE
Grok 4.20—
Qwen3.5 397B28.7%
Source
Not directly comparable

Math

AIME26
Grok 4.20—
Qwen3.5 397B93.3%
Source
Not directly comparable
HMMT Feb 2025
Grok 4.20—
Qwen3.5 397B94.8%
Source
Not directly comparable
HMMT Nov 2025
Grok 4.20—
Qwen3.5 397B92.7%
Source
Not directly comparable
HMMT Feb 2026
Grok 4.20—
Qwen3.5 397B87.9%
Source
Not directly comparable
MMAnswerBench
Grok 4.20—
Qwen3.5 397B80.9%
Source
Not directly comparable

Multilingual

MMLU-ProX
Grok 4.20—
Qwen3.5 397B84.7%
Source
Not directly comparable
NOVA-63
Grok 4.20—
Qwen3.5 397B59.1%
Source
Not directly comparable

Multimodal

MMMU-Pro
Grok 4.2075.2%
Source
Qwen3.5 397B79%
Source
Qwen3.5 397B leads this result
CharXiv
Grok 4.2060.9%
Source
Qwen3.5 397B80.8%
Source
Qwen3.5 397B leads this result
ERQA
Grok 4.2054.1%
Source
Qwen3.5 397B—
Not directly comparable
SimpleVQA
Grok 4.2057.4%
Source
Qwen3.5 397B—
Not directly comparable
MedXpertQA (MM)
Grok 4.2065.8%
Source
Qwen3.5 397B—
Not directly comparable
MathVision
Grok 4.20—
Qwen3.5 397B88.6%
Source
Not directly comparable
VideoMMMU
Grok 4.20—
Qwen3.5 397B84.7%
Source
Not directly comparable
ScreenSpot Pro
Grok 4.20—
Qwen3.5 397B65.6%
Source
Not directly comparable
V*
Grok 4.20—
Qwen3.5 397B95.8%
Source
Not directly comparable

Instruction following

IFEval
Grok 4.20—
Qwen3.5 397B92.6%
Source
Not directly comparable

Frequently asked questions

Which is better, Grok 4.20 or Qwen3.5 397B?

Grok 4.20 has the higher public score estimate, 67.13 versus 54.7, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.

Which is better for coding, Grok 4.20 or Qwen3.5 397B?

Qwen3.5 397B scores higher for coding on the public lane, 50.2 to 28.2. Qwen3.5 397B is scored on Estimated evidence for coding, so the reading is directional rather than like-for-like. BenchLM does not name a winner for coding from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which is better for agentic tasks, Grok 4.20 or Qwen3.5 397B?

Qwen3.5 397B scores higher for agentic tasks on the public lane, 46.8 to 26.7. Qwen3.5 397B is scored on Estimated evidence for agentic tasks, so the reading is directional rather than like-for-like. BenchLM does not name a winner for agentic tasks from a directional reading; read the shared benchmark rows directly and test the models on the same task set.

Which costs less, Grok 4.20 or Qwen3.5 397B?

For the stated presets, chat costs $0.005 on Grok 4.20 and $0.0024 on Qwen3.5 397B; repository review costs $0.118 and $0.0408; the cache-heavy agent loop costs $0.5 and $0.168. Qwen3.5 397B does not fit this workload in one request. Grok 4.20 has no published cached-input rate, so cached tokens use its listed input rate. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate.

Which has the larger context window, Grok 4.20 or Qwen3.5 397B?

Grok 4.20 has the larger documented context window: 2M, compared with 128K.

Related comparisons

Compare API pricing Read the methodology Open the model selector

Last updated September 10, 2026

Watch Grok 4.20 vs Qwen3.5 397B

One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.

Read a sample issue

Join 2,000+ readers.

Grok 4.20 vs Qwen3.5 397B

Which one for your work

Long documents

Chat turn cost

Repository review cost

Coding work

Agentic work

Cache-heavy agent loop cost

What is actually comparable

Category results, on a stated basis

Multimodal

Agentic

Coding

Knowledge

Reasoning

Math

Multilingual

Instruction following

Shape of the matched evidence

What each workload costs

Chat turn

Repository review

Cache-heavy agent loop

Specification differences

Context window

API model ID

Cached-input rate

Documented inputs

Documented outputs

Provider availability

Reasoning profile

Weight access

License

Release date

Benchmark evidence

Agentic

Coding

Reasoning

Knowledge

Math

Multilingual

Multimodal

Instruction following

Frequently asked questions

Which is better, Grok 4.20 or Qwen3.5 397B?

Which is better for coding, Grok 4.20 or Qwen3.5 397B?

Which is better for agentic tasks, Grok 4.20 or Qwen3.5 397B?

Which costs less, Grok 4.20 or Qwen3.5 397B?

Which has the larger context window, Grok 4.20 or Qwen3.5 397B?

Related comparisons

Watch Grok 4.20 vs Qwen3.5 397B