Long documents
Prompts that approach the documented context limit
Spark-X2.6-4B
Spark-X2.6-4B has the larger documented context window.
Confidence: documented
Qwen3.5 397B vs Spark-X2.6-4B
Updated October 20, 2026. Rank says Qwen3.5 397B is ahead. Price, access, and your workload can each overturn that. Public scores include evidence status and uncertainty.
Share or export
Decision reading
Qwen3.5 397B has the higher public point estimate, 55.39 versus 40.14. Their conditional score ranges overlap. These ranges do not establish rank confidence. 13 results are shared. Category rows resting on Estimated evidence or different benchmark sets are marked directional and do not name a winner.
- Shared results
- 13
- Qwen3.5 397B only
- 49
- Spark-X2.6-4B only
- 7
- Like-for-like categories
- 1 / 8
Estimated: Qwen3.5 397B and Spark-X2.6-4B. Conditional ranges do not establish rank confidence.How the comparison works
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.
Show secondary and unsupported calls
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
Qwen3.5 397B is not ranked on the public lane for coding, so no winner is named for coding.
Confidence: limitedAgentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
Qwen3.5 397B and Spark-X2.6-4B are scored on Estimated evidence for agentic, so the reading is directional rather than like-for-like.
Confidence: limitedChat turn cost
1K fresh input + 500 output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-ratesCache-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. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate. Spark-X2.6-4B has no comparable published API token rate.
Confidence: rate-fallbackRepository review cost
50K fresh input + 3K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
Which one for a specific job
Choose a job from the LLM Selector's task catalog to see the category row it rests on, under the same basis rules as the table below. A directional row stays directional; choosing a job never creates a winner.
The same task catalog as the LLM Selector. Each job names the evidence surface it rests on; nothing here adds a new score.
—Qwen3.5 397B30.6Spark-X2.6-4B
Not comparable · BenchAlign v5.8
The coding row is not comparable on the public lane: at least one model is not measured or not ranked there.
Coding scores combine specific tasks and setups. Match the editor, harness, and effort to your workflow.
Same basis rules as the category table below
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
49 Qwen3.5 397B only
13 Shared
7 Spark-X2.6-4B only
2 categories rest on Estimated evidence or different evidence sets. Those rows remain visible for coverage context but do not name a winner.
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.
GPQAKnowledge
Normalized gap 18.5Qwen3.5 397B:88.4%Spark-X2.6-4B:69.9%BrowseCompAgentic
Normalized gap 15.9Qwen3.5 397B:69%Spark-X2.6-4B:53.1%SWE-bench VerifiedCoding
Normalized gap 13.4Qwen3.5 397B:76.2%Spark-X2.6-4B:62.8%LongBench v2Reasoning
Normalized gap 9.9Qwen3.5 397B:63.2%Spark-X2.6-4B:53.3%HMMT Feb 2026Math
Normalized gap 6.6Qwen3.5 397B:87.9%Spark-X2.6-4B:81.3%
Bars run 0–100 on each benchmark’s normalized display scale
Category results, on a stated basis
Each row shows the public-lane category score for both models: the BenchAlign v5.8 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.
Instruction following
Like-for-like- Qwen3.5 397B
- 79.9
- #59/149
- Spark-X2.6-4B
- 78.2
- #62/149
- Basis
- Provisional lane · 1 vs 1 weighted rows
- Reading
- Qwen3.5 397B leads
Agentic
Directional only- Qwen3.5 397B
- 44.2
- Estimated · #62/131
- Spark-X2.6-4B
- 33.0
- Estimated · #89/131
- Basis
- BenchAlign v5.8 lane · 14 vs 7 public rows
- Reading
- Directional only
Knowledge
Directional only- Qwen3.5 397B
- 53.2
- Estimated · #70/187
- Spark-X2.6-4B
- 43.0
- Estimated · #109/187
- Basis
- BenchAlign v5.8 lane · 6 vs 3 public rows
- Reading
- Directional only
Coding
Not comparable- Qwen3.5 397B
- Not ranked
- Spark-X2.6-4B
- 30.6
- Estimated · #95/152
- Basis
- BenchAlign v5.8 lane · 6 vs 4 public rows
- Reading
- Not comparable
Reasoning
Not comparable- Qwen3.5 397B
- 74.1
- Unranked · 2 rankable rows
- Spark-X2.6-4B
- 48.8
- Unranked · 2 rankable rows
- Basis
- Provisional lane · 1 vs 1 weighted rows
- Reading
- Not comparable
Multimodal
Not comparable- Qwen3.5 397B
- 73.6
- #29/55
- Spark-X2.6-4B
- Not ranked
- Basis
- Provisional lane · 2 vs 0 weighted rows
- Reading
- Not comparable
Multilingual
Not comparable- Qwen3.5 397B
- 93.8
- #5/17
- Spark-X2.6-4B
- Not ranked
- Basis
- Provisional lane · 1 vs 0 weighted rows
- Reading
- Not comparable
Math
Not comparable- Qwen3.5 397B
- 81.7
- Unranked · 5 rankable rows
- Spark-X2.6-4B
- 76.4
- Unranked · 3 rankable rows
- Basis
- Provisional lane · 2 vs 2 weighted rows
- Reading
- Not comparable
| Category | Qwen3.5 397B | Spark-X2.6-4B | Basis | Reading |
|---|---|---|---|---|
| Instruction following | 79.9#59/149 | 78.2#62/149 | Like-for-likeProvisional lane · 1 vs 1 weighted rows | Qwen3.5 397B leads |
| Agentic | 44.2Estimated · #62/131 | 33.0Estimated · #89/131 | Directional onlyBenchAlign v5.8 lane · 14 vs 7 public rows | Directional only |
| Knowledge | 53.2Estimated · #70/187 | 43.0Estimated · #109/187 | Directional onlyBenchAlign v5.8 lane · 6 vs 3 public rows | Directional only |
| Coding | Not ranked | 30.6Estimated · #95/152 | Not comparableBenchAlign v5.8 lane · 6 vs 4 public rows | Not comparable |
| Reasoning | 74.1Unranked · 2 rankable rows | 48.8Unranked · 2 rankable rows | Not comparableProvisional lane · 1 vs 1 weighted rows | Not comparable |
| Multimodal | 73.6#29/55 | Not ranked | Not comparableProvisional lane · 2 vs 0 weighted rows | Not comparable |
| Multilingual | 93.8#5/17 | Not ranked | Not comparableProvisional lane · 1 vs 0 weighted rows | Not comparable |
| Math | 81.7Unranked · 5 rankable rows | 76.4Unranked · 3 rankable rows | Not comparableProvisional lane · 2 vs 2 weighted rows | Not comparable |
Ranks count the models scored in each category’s lane, so the agentic, coding, and knowledge denominators (BenchAlign v5.8) differ from the provisional-lane categories. Unranked scores sit on the provisional lane’s scale but never name a winner. Like-for-like rows within 0.5 points read as a practical tie.
Bars run 0–100Methodology
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
- Qwen3.5 397B
- $0.0024
- Fits in one request
- Spark-X2.6-4B
- Self-hosted; infrastructure cost varies
- Fits in one request
Spark-X2.6-4B has no comparable published API token rate.
Repository review
50K fresh input + 3K output tokens
- Qwen3.5 397B
- $0.0408
- Fits in one request
- Spark-X2.6-4B
- Self-hosted; infrastructure cost varies
- Fits in one request
Spark-X2.6-4B has no comparable published API token rate.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- Qwen3.5 397B
- $0.168
- Does not fit in one request
- Cached input priced at the published list-input rate
- Spark-X2.6-4B
- Self-hosted; infrastructure cost varies
- Fits in one request
- Cached-input rate unavailable
Qwen3.5 397B does not fit this workload in one request. Qwen3.5 397B has no published cached-input rate, so cached tokens use its listed input rate. Spark-X2.6-4B has no comparable published API token rate.
Cached input falls back to the list input rate only where a cached rate is unpublished
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationQwen3.5 397BSpark-X2.6-4B
Context window
Maximum documented context; output-token limits may be lower.
Qwen3.5 397B
128K
Spark-X2.6-4B
API model ID
Qwen3.5 397B
Not sourced
Spark-X2.6-4B
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
Qwen3.5 397B
Not published
Spark-X2.6-4B
No comparable hosted API rate
SparkLLM Spark-X2.6-4B model cardDocumented inputs
Qwen3.5 397B
Not sourced
Spark-X2.6-4B
Documented outputs
Qwen3.5 397B
Not sourced
Spark-X2.6-4B
Provider availability
Qwen3.5 397B
Not sourced
Spark-X2.6-4B
Open Weights · Hugging Face, self-hosted inference
SparkLLM Spark-X2.6-4B model cardReasoning profile
Qwen3.5 397B
Non-Reasoning
Spark-X2.6-4B
Reasoning
Weight access
Qwen3.5 397B
Open Weight
Spark-X2.6-4B
Open Weight
License
Qwen3.5 397B
Open Weight
Spark-X2.6-4B
Open Weight
Release date
Qwen3.5 397B
2026-02-16
Spark-X2.6-4B
2026-10-19
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
- Qwen3.5 397B has the higher public point estimate, 55.39 versus 40.14. Their conditional score ranges overlap. These ranges do not establish rank confidence.
- Workload cost
- A complete comparable API-rate estimate is not available for both models.
- Context tradeoff
- Spark-X2.6-4B has the larger documented window (1M).
Run the same representative tasks against both endpoints before changing production traffic.
Questions
Which is better, Qwen3.5 397B or Spark-X2.6-4B?
Qwen3.5 397B has the higher public point estimate, 55.39 versus 40.14. Their conditional score ranges overlap. These ranges do not establish rank confidence. The page therefore keeps the decision tied to the specific documented workload.
Which is better for coding, Qwen3.5 397B or Spark-X2.6-4B?
Qwen3.5 397B is not ranked on the public lane for coding, so no winner is named for coding.
Which is better for agentic tasks, Qwen3.5 397B or Spark-X2.6-4B?
Qwen3.5 397B scores higher for agentic tasks on the public lane, 44.2 to 33. Qwen3.5 397B and Spark-X2.6-4B are 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, Qwen3.5 397B or Spark-X2.6-4B?
Both models do not have comparable published API token rates, so this page does not name a universal price winner.
Which has the larger context window, Qwen3.5 397B or Spark-X2.6-4B?
Spark-X2.6-4B has the larger documented context window: 1M, compared with 128K.
Self-host vs API cost
Estimates at 50,000 req/day · 1000 tokens/req average.
Qwen3.5 397B
API / mo$3,150
Self-host / moNot listed
Break-even—
Proprietary model — self-hosting not applicable.
Spark-X2.6-4B
API / mo$0
Self-host / mo$298
Break-even—
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence69 rows
Agentic
Terminal-Bench 2.0
Qwen3.5 397B52.5%SourceSpark-X2.6-4B—Not directly comparable
BrowseComp
Qwen3.5 397B69%SourceSpark-X2.6-4B53.1%SourceQwen3.5 397B leads this result
OSWorld-Verified
Qwen3.5 397B62.2%SourceSpark-X2.6-4B—Not directly comparable
Claw-Eval
Qwen3.5 397B56.8%SourceSpark-X2.6-4B69.9%SourceSpark-X2.6-4B leads this result
QwenClawBench
Qwen3.5 397B51.8%SourceSpark-X2.6-4B—Not directly comparable
τ³-bench results
Qwen3.5 397B68.4%SourceSpark-X2.6-4B34.7%SourceQwen3.5 397B leads this result
VITA-Bench
Qwen3.5 397B43.7%SourceSpark-X2.6-4B—Not directly comparable
DeepPlanning
Qwen3.5 397B37.6%SourceSpark-X2.6-4B—Not directly comparable
Toolathlon
Qwen3.5 397B36.3%SourceSpark-X2.6-4B—Not directly comparable
MCP Atlas
Qwen3.5 397B46.1%SourceSpark-X2.6-4B62.4%SourceSpark-X2.6-4B leads this result
MCP-Tasks
Qwen3.5 397B74.2%SourceSpark-X2.6-4B—Not directly comparable
WideResearch
Qwen3.5 397B74.0%SourceSpark-X2.6-4B—Not directly comparable
Gert Labs
Qwen3.5 397B46.76%SourceSpark-X2.6-4B—Not directly comparable
ResearchClawBench
Qwen3.5 397B14.2%SourceSpark-X2.6-4B—Not directly comparable
BFCL v4
Qwen3.5 397B—Spark-X2.6-4B71.8%SourceNot directly comparable
τ²-bench results
Qwen3.5 397B—Spark-X2.6-4B86.9%SourceNot directly comparable
HLE w/ tools
Qwen3.5 397B—Spark-X2.6-4B36.9%SourceNot directly comparable
Coding
SWE-bench Verified
Qwen3.5 397B76.2%SourceSpark-X2.6-4B62.8%SourceQwen3.5 397B leads this result
LiveCodeBench v6
Qwen3.5 397B83.6%SourceSpark-X2.6-4B—Not directly comparable
SWE-bench Pro
Qwen3.5 397B50.9%SourceSpark-X2.6-4B54.3%SourceSpark-X2.6-4B leads this result
SWE Multilingual
Qwen3.5 397B69.3%SourceSpark-X2.6-4B64.7%SourceQwen3.5 397B leads this result
NL2Repo
Qwen3.5 397B32.2%SourceSpark-X2.6-4B—Not directly comparable
Terminal-Bench 2.0
Qwen3.5 397B52.5%SourceSpark-X2.6-4B—Not directly comparable
SciCode
Qwen3.5 397B—Spark-X2.6-4B36.8%SourceNot directly comparable
Reasoning
Multimodal
MMMU
Qwen3.5 397B85.0%SourceSpark-X2.6-4B—Not directly comparable
MMMU-Pro
Qwen3.5 397B79%SourceSpark-X2.6-4B—Not directly comparable
RealWorldQA
Qwen3.5 397B83.9%SourceSpark-X2.6-4B—Not directly comparable
OmniDocBench 1.5
Qwen3.5 397B90.8%SourceSpark-X2.6-4B—Not directly comparable
Video-MME (with subtitle)
Qwen3.5 397B87.5%SourceSpark-X2.6-4B—Not directly comparable
Video-MME (w/o subtitle)
Qwen3.5 397B83.7%SourceSpark-X2.6-4B—Not directly comparable
MathVision
Qwen3.5 397B88.6%SourceSpark-X2.6-4B—Not directly comparable
We-Math
Qwen3.5 397B87.9%SourceSpark-X2.6-4B—Not directly comparable
DynaMath
Qwen3.5 397B86.3%SourceSpark-X2.6-4B—Not directly comparable
SimpleVQA
Qwen3.5 397B67.1%SourceSpark-X2.6-4B—Not directly comparable
CharXiv
Qwen3.5 397B80.8%SourceSpark-X2.6-4B—Not directly comparable
MMLongBench-Doc
Qwen3.5 397B61.5%SourceSpark-X2.6-4B—Not directly comparable
CC-OCR
Qwen3.5 397B82.0%SourceSpark-X2.6-4B—Not directly comparable
AI2D_TEST
Qwen3.5 397B93.9%SourceSpark-X2.6-4B—Not directly comparable
CountBench
Qwen3.5 397B97.2%SourceSpark-X2.6-4B—Not directly comparable
RefCOCO (avg)
Qwen3.5 397B92.3%SourceSpark-X2.6-4B—Not directly comparable
ODINW13
Qwen3.5 397B47.0%SourceSpark-X2.6-4B—Not directly comparable
ERQA
Qwen3.5 397B67.5%SourceSpark-X2.6-4B—Not directly comparable
VideoMMMU
Qwen3.5 397B84.7%SourceSpark-X2.6-4B—Not directly comparable
MLVU (M-Avg)
Qwen3.5 397B86.7%SourceSpark-X2.6-4B—Not directly comparable
ScreenSpot Pro
Qwen3.5 397B65.6%SourceSpark-X2.6-4B—Not directly comparable
V*
Qwen3.5 397B95.8%SourceSpark-X2.6-4B—Not directly comparable
Knowledge
GPQA
Qwen3.5 397B88.4%SourceSpark-X2.6-4B69.9%SourceQwen3.5 397B leads this result
SuperGPQA
Qwen3.5 397B70.4%SourceSpark-X2.6-4B—Not directly comparable
MMLU-Pro
Qwen3.5 397B87.8%SourceSpark-X2.6-4B—Not directly comparable
MMLU-Redux
Qwen3.5 397B94.9%SourceSpark-X2.6-4B—Not directly comparable
C-Eval
Qwen3.5 397B93%SourceSpark-X2.6-4B—Not directly comparable
HLE
Qwen3.5 397B28.7%SourceSpark-X2.6-4B—Not directly comparable
GPQA-D
Qwen3.5 397B—Spark-X2.6-4B77.1%SourceNot directly comparable
HLE w/o tools
Qwen3.5 397B—Spark-X2.6-4B17.9%SourceNot directly comparable
Multilingual
MMLU-ProX
Qwen3.5 397B84.7%SourceSpark-X2.6-4B—Not directly comparable
NOVA-63
Qwen3.5 397B59.1%SourceSpark-X2.6-4B—Not directly comparable
INCLUDE
Qwen3.5 397B85.6%SourceSpark-X2.6-4B—Not directly comparable
PolyMath
Qwen3.5 397B73.3%SourceSpark-X2.6-4B—Not directly comparable
MAXIFE
Qwen3.5 397B88.2%SourceSpark-X2.6-4B—Not directly comparable
Instruction following
Math
AIME26
Qwen3.5 397B93.3%SourceSpark-X2.6-4B91.2%SourceQwen3.5 397B leads this result
HMMT Feb 2025
Qwen3.5 397B94.8%SourceSpark-X2.6-4B—Not directly comparable
HMMT Nov 2025
Qwen3.5 397B92.7%SourceSpark-X2.6-4B—Not directly comparable
HMMT Feb 2026
Qwen3.5 397B87.9%SourceSpark-X2.6-4B81.3%SourceQwen3.5 397B leads this result
MMAnswerBench
Qwen3.5 397B80.9%SourceSpark-X2.6-4B—Not directly comparable
IMOAnswerBench
Qwen3.5 397B—Spark-X2.6-4B75.2%SourceNot directly comparable
69 public results · 13 shared
Watch Qwen3.5 397B vs Spark-X2.6-4B
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 5,600+ readers.
Last updated October 20, 2026