vs
3:2
overall capability wins (0 ties across 5 scored). Atria Dawn Preview leads overall.
- Coding1:0
- Tool use1:0
- Specs1:2
Verdict
Atria Dawn Preview leads coding (widest gap: +3.7 on SWE-Pro); Inkling-Small offers ~4.1x larger context (1.0M vs 256K); Inkling-Small allows ~2.0x more max output (131K vs 66K).
- Atria Dawn Preview leads coding (widest gap: +3.7 on SWE-Pro)
- Inkling-Small offers ~4.1x larger context (1.0M vs 256K)
- Inkling-Small allows ~2.0x more max output (131K vs 66K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Atria Dawn Preview
- -
- Inkling-Small
- $0.94
Price per 1M tokens
Full comparison
Organization
Shanghai AI Lab
Thinking Machines
Family
Other
Other
License
MIT
Apache 2.0
Open weights
Yes
Yes
Release
Sep 11, 2026
Jul 30, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
Tinker (ref.)
Modalities
text → text
text, image, audio → text
Specs
Context window
256K
1.0M
Max output
66K
131K
Parameters
744B
276B (12B act.)
Pricing
Input $/1M
—
$0.58
Output $/1M
—
$1.44
Blended $/1M (3∶1)
-
$0.79
Speed
tok/s
-
85
TTFT (s)
-
0.35
Reasoning
MMLU-Pro
—
—
GPQA Diamond
—
89.5
Humanity's Last Exam
—
33.3
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
33.2
AA-LCR v1.1
—
75.7
CritPt
—
8.3
MMMU-Pro
—
74.0
IFBench
—
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
80.2
SWE-bench Pro
59.6
55.9
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
55.1
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
—
49.7
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,407
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,405
| Metric | Atria Dawn Preview | Inkling-Small | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Shanghai AI Lab | Thinking Machines | — |
| Family | Other | Other | — |
| License | MIT | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Sep 11, 2026 | Jul 30, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | Tinker (ref.) | — |
| Modalities | text → text | text, image, audio → text | — |
| Specs | |||
| Context window | 256K | 1.0M | B +793K |
| Max output | 66K | 131K | B +66K |
| Parameters | 744B | 276B (12B act.) | A +468 |
| Pricing | |||
| Input $/1M | — | $0.58 | — |
| Output $/1M | — | $1.44 | — |
| Blended $/1M (3∶1) | - | $0.79 | — |
| Speed | |||
| tok/s | - | 85 | — |
| TTFT (s) | - | 0.35 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | — | 89.5 | — |
| Humanity's Last Exam | — | 33.3 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 33.2 | — |
| AA-LCR v1.1 | — | 75.7 | — |
| CritPt | — | 8.3 | — |
| MMMU-Pro | — | 74.0 | — |
| IFBench | — | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | 80.2 | — |
| SWE-bench Pro | 59.6 | 55.9 | A +3.7 pts |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 55.1 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | — | 49.7 | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,407 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,405 | — |
Atria Dawn Preview
- Context
- 256K / 66K out
- Parameters
- 744B
- Price
- —
- Speed
- —
- Modalities
- text → text
- License
- MIT
Shanghai AI Laboratory's open-weight agentic MoE (744B total, MIT license, 256K context), built on a GLM-5.2 base and trained by the cross-institutional ATRIA initiative (with Fudan NLP Lab) for long-horizon research and engineering workflows: problem analysis, tool use, code, experiments, and failure recovery. Benchmark scores are from the model's own technical report and have not yet been independently verified by a third-party evaluator.
Inkling-Small
- Context
- 1.0M / 131K out
- Parameters
- 276B (12B act.)
- Price
- $0.58 / $1.44
- Speed
- 85 tok/s · 0.35s TTFT
- Modalities
- text, image, audio → text
- License
- Apache 2.0
Efficient Inkling sibling (276B / 12B active MoE) — matches or beats Inkling on many benchmarks at ~¼ the size; Apache 2.0 weights with Tinker fine-tuning access.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.