Across 25 comparable benchmarks, Inkling 2 Small has better reported results on 21 and worse on 4. 3 further shared results are listed below, each with the reason it is not counted.
Moderate evidence25 of 28 shared results countedHow we compare
Price basis: Standard rates vs listed rates. The price section below shows the applicable periods.
| Fact | Inkling 2 Small | Qwen3.8-27B |
|---|---|---|
| Context window | 1M tokens | 256K tokens |
| Input / 1M | $0.300 | $0.500 |
| Output / 1M | $1.00 | $3.00 |
| Blended (4:1) | $0.246 | $0.712 |
| Status | Available | Available |
| Released | November 5, 2026 | August 14, 2026 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 16 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Inkling 2 Small ahead on 4 of 4 counted
Inkling 2 Small ahead on 1 of 1 counted
Inkling 2 Small ahead on 1 of 1 counted
Inkling 2 Small ahead on 2 of 3 counted
Inkling 2 Small ahead on 1 of 1 counted
No fixed-scale average — see the rows in Benchmark results
Inkling 2 Small ahead on 1 of 1 counted
Inkling 2 Small ahead on 5 of 7 counted
Inkling 2 Small ahead on 4 of 5 counted
No shared comparable results for Mathematics, Cybersecurity.
Published scores, with matching modes where available. Gaps show Inkling 2 Small relative to Qwen3.8-27B. Reading the results
| Benchmark | Inkling 2 Small | Qwen3.8-27B | Gap |
|---|---|---|---|
| DeepSWERepository Engineering · Pass@1 (DeepSWE v1.1) | 72.1Extra-High · tools | 42.2≠Thinking · tools | +29.9 pts#20 vs #64 of 73 |
| Terminal-Bench 2.1Agentic Development · Accuracy | 85.1Extra-High · tools | 73.0≠Thinking · tools | +12.1 pts#24 vs #46 of 68 |
| Terminal-Bench 4.0Agentic Development · Resolution rate (%) | 13.1Extra-High · tools | 5.6Extra-High · tools | +7.5 pts#57 vs #67 of 90 |
| Vibe Code Bench v1.1Code Generation & Editing · Pass rate (%) | 81.6Extra-High · tools | 64.8Extra-High · tools | +16.8 pts#30 vs #54 of 121 |
| ARC-AGI-1Abstract Generalization · Score (%) | 96.5Extra-High · no tools | 87.5Extra-High · no tools | +9.0 pts#13 vs #40 of 87 |
| ARC-AGI-2Abstract Generalization · Score (% solved); cost per task (USD)Not counted · Speed and cost telemetry is reported, not scored | 81.7Extra-High · no tools | 42.4Extra-High · no tools | 1.93×#21 vs #48 of 80 |
| GPQA DiamondScientific Reasoning · Accuracy | 93.0Extra-High · no tools | 89.2≠Thinking · no tools | +3.8 pts#35 vs #58 of 200 |
| AutomationBench-AATool Orchestration · Guardrail-safe objective completion rate | 45.6Extra-High · tools | 48.2≠Thinking · tools | −2.6 pts#16 vs #15 of 18 |
| MCP-AtlasTool Orchestration · Pass rate / claim coverage | 85.4Extra-High · tools | 77.3≠Thinking · tools | +8.1 pts#7 vs #25 of 55 |
| ScreenSpot-Pro (no tools)GUI Grounding · Success rate (%) | 83.8Extra-High · no tools | 57.6≠Thinking · no tools | +26.2 pts#2 vs #6 of 8 |
| DeepSearchQAFact Finding · F1 score | 90.9Extra-High · tools | 79.7≠Thinking · tools | 1.14×#2 vs #6 of 9 |
| LongBench v2Long Reasoning · Accuracy | 70.6Extra-High · no tools | 52.3≠Thinking · no tools | +18.3 pts#1 vs #17 of 22 |
| BabyVisionVisual Understanding · Accuracy | 93.7Extra-High · tools | 85.6≠Thinking · tools | +8.1 pts#7 vs #13 of 17 |
| ChartographyDocuments & Charts · Score | 80.4Extra-High · tools | 52.1≠Thinking · tools | 1.54×#10 vs #24 of 29 |
| CharXiv RQDocuments & Charts · Score | 87.7Extra-High · tools | 90.2≠Thinking · tools | 0.97×#11 vs #7 of 21 |
| GDP.pdfDocuments & Charts · Task pass rate (%) | 25.2Extra-High · no tools | 16.4Extra-High · no tools | +8.8 pts#17 vs #39 of 96 |
| MathVisionVisual Understanding · Accuracy | 92.8Extra-High · no tools | 94.6≠Thinking · tools | −1.8 pts#5 vs #3 of 10 |
| MMMU-ProVisual Understanding · Accuracy | 81.6Extra-High · no tools | 76.3Extra-High · no tools | +5.3 pts#25 vs #58 of 136 |
| PerceptionBenchVisual Understanding · Accuracy | 64.7Extra-High · no tools | 52.4≠Thinking · no tools | +12.3 pts#6 vs #14 of 19 |
| AA-BriefcaseOffice & Business · AA-Briefcase Elo | 1527Extra-High · tools | 1391≠Thinking · tools | 1.10×#15 vs #17 of 20 |
| GDPval-AA v2Cross-industry Work · Elo score | 1630Extra-High · tools | 1427≠Thinking · tools | 1.14×#14 vs #17 of 23 |
| JobBenchCross-industry Work · Mean task reward (%)Not counted · This metric has no better direction | 70.5Extra-High · tools | 33.4≠Thinking · tools | — |
| Legal Research BenchLegal · Score | 46.6Extra-High · tools | 36.1Extra-High · tools | 1.29×#18 vs #49 of 91 |
| Tax Agent BenchFinance · Accuracy (%, rubric × citation quality) | 65.8Extra-High · tools | 30.9Extra-High · tools | +34.9 pts#10 vs #41 of 74 |
| τ³-BankingService Workflows · Score | 40.4Extra-High · tools | 48.0Extra-High · tools | 0.84×#20 vs #7 of 120 |
| HLEKnowledge Exams · AccuracyNot counted · Different or unrecorded test environments — excluded from the tally | 44.1Extra-High · no tools · text only· text-only | 30.8≠Thinking · no tools· legacy-default | —#69 vs #86 of 151 |
| HLE-DiamondKnowledge Exams · Accuracy (%) | 28.7Extra-High · no tools | 15.3≠Thinking · no tools | +13.4 pts#9 vs #12 of 16 |
| AA Intelligence Index v5.0.1Capability Indices · Index (0-100) | 30.6Extra-High · tools | 20.1Extra-High · tools | 1.52×#30 vs #43 of 62 |
Repository Engineering · Pass@1 (DeepSWE v1.1)
Agentic Development · Accuracy
Agentic Development · Resolution rate (%)
Code Generation & Editing · Pass rate (%)
Scientific Reasoning · Accuracy
Tool Orchestration · Guardrail-safe objective completion rate
Tool Orchestration · Pass rate / claim coverage
GUI Grounding · Success rate (%)
Fact Finding · F1 score
Long Reasoning · Accuracy
Visual Understanding · Accuracy
Documents & Charts · Score
Documents & Charts · Score
Documents & Charts · Task pass rate (%)
Visual Understanding · Accuracy
Visual Understanding · Accuracy
Visual Understanding · Accuracy
Office & Business · AA-Briefcase Elo
Cross-industry Work · Elo score
Cross-industry Work · Mean task reward (%)
Not counted · This metric has no better direction
Legal · Score
Finance · Accuracy (%, rubric × citation quality)
Service Workflows · Score
Knowledge Exams · Accuracy
Not counted · Different or unrecorded test environments — excluded from the tally
Knowledge Exams · Accuracy (%)
Capability Indices · Index (0-100)
A dash means no result is available.
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix · 5% steps
Blended price = effective input rate at 90% cache hits × 80% + output rate × 20%
Inkling 2 Small costs66%lessthan Qwen3.8-27B per million tokens — 2.9× cheaper.Using standard rates vs listed rates at this token mix.
Inkling 2 Small: Standard pricing. Tinker 64K/256K inference; training billed separately. Image-specific token accounting not separately specified.
Qwen3.8-27B: International pricing. context_length <= 1000000 Alibaba Model Studio International (Singapore)
| Price / 1M tokens | Inkling 2 Small | Qwen3.8-27B | Ratio |
|---|---|---|---|
| Input | $0.30 | $0.50 | 0.60× |
| Output | $1.00 | $3.00 | 0.33× |
| Cached input | $0.03 | $0.10 | 0.30× |
| Cache write | — | — | — |
| Blended | $0.246 | $0.712 | 0.34× |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
Inkling 2 Small relative to Qwen3.8-27B
Summary: 3 of 3 shared benchmarks have matching recorded settings and sources. The average is a geometric mean of 3 positive-cost ratios, one equal weight per benchmark. This benchmark set can include related tests and indexes; it is not a workload cost estimate.
Inkling 2 Small 3× cheaper
Inkling 2 Small 2.6× cheaper
Inkling 2 Small 1.6× cheaper
Inkling 2 Small 3× cheaper
Inkling 2 Small 2.6× cheaper
Inkling 2 Small 1.6× cheaper
Effort, protocol, aggregation and source match; other run conditions may differ.
Inkling 2 Small
Qwen3.8-27B
Effort, protocol, aggregation and source match; other run conditions may differ.
Inkling 2 Small
Qwen3.8-27B
Effort, protocol, aggregation and source match; other run conditions may differ.
Inkling 2 Small
Qwen3.8-27B
| Benchmark | Inkling 2 Small | Qwen3.8-27B |
|---|---|---|
| AA-Briefcase | Not reported | |
| AA-Omniscience | Not reported | |
| AutomationBench-AA | $0.0581 / task Average · Guardrail-safe objective completion rate: 45.6 Extra-High (With Tools) | Not reported |
| FrontierCS 2 | Not reported | |
| GDP.pdf | Not reported | |
| HLE-Diamond | Not reported | |
| LiveBench 2.0 | Not reported | $0.519 / task Average · Overall score (equal-weight mean of 7 suite scores, 0-100): 62.7 Max (task-specific tools) |
| Tax Agent Bench | Not reported | $0.8153 / task Average · Accuracy (%, rubric × citation quality): 30.9 Extra-High (With Tools) |
| Terminal-Bench 4.0 | $8.22 / task Average · Resolution rate (%): 12.6 Unspecified Reasoning (With Tools) · vals-mini-swe | Not reported |
| Terminal-Bench 4.1 | Not reported | |
| Terminal-Bench-Science 0.1 | Not reported |
| Attribute | Inkling 2 Small | Qwen3.8-27B |
|---|---|---|
| Availability | Available | Available |
| Provider | Thinking Machines Lab | Alibaba |
| Model type | Reasoning model | Reasoning model |
| First release | November 5, 2026 | August 14, 2026 |
| Context window | 1M tokens | 256K tokens |
| Max output | — | 128K tokens |
| Total parameters | 594B | 27B |
| Active parameters | 17B | 27B |
| Architecture | Mixture of Experts (MoE) | Dense |
| Documented capabilities | Reasoning, Coding, Tool use, Long context | Reasoning, Multilingual, Coding, Tool use, Function calling, Long context, Computer use |
| Input modalities | Text, Image | Text, Image, Video |
| Output modalities | Text | Text |
| Weights | Open weights | Open weights |
| Licence | Apache 2.0 | Apache 2.0 |
| Commercial use | Commercial use allowed | Commercial use allowed |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used Inkling 2 Small or Qwen3.8-27B? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.