vs
0:6
overall capability wins (0 ties across 6 scored). Inkling leads overall.
- Reasoning0:3
- Coding0:1
- Specs0:2
Verdict
Inkling leads coding (widest gap: +8.9 on SciCode); Inkling offers ~4.0x larger context (1.0M vs 262K).
- Inkling leads coding (widest gap: +8.9 on SciCode)
- Inkling offers ~4.0x larger context (1.0M vs 262K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Agnes-3.0-Flash Preview
- -
- Inkling
- $2.01
Price per 1M tokens
Full comparison
Organization
Agnes AI
Thinking Machines
Family
Other
Other
License
Apache 2.0
Apache 2.0
Open weights
Yes
Yes
Release
Sep 13, 2026
Jul 15, 2026
Knowledge cutoff
-
-
API / provider
No primary API price listed
OpenRouter / Tinker (ref.)
Modalities
text, image, video → text
text, image, audio → text
Specs
Context window
262K
1.0M
Max output
-
131K
Parameters
33B
975B (41B act.)
Pricing
Input $/1M
—
$1
Output $/1M
—
$4.05
Blended $/1M (3∶1)
-
$1.76
Speed
tok/s
-
55
TTFT (s)
-
0.55
Reasoning
MMLU-Pro
—
—
GPQA Diamond
85.0
87.2
Humanity's Last Exam
—
31.9
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
23.0
41.6
AA-LCR v1.1
68.3
77.3
CritPt
—
5.4
MMMU-Pro
—
73.5
IFBench
74.2
—
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
77.6
SWE-bench Pro
—
54.3
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
55.1
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
38.1
47.0
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,410
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,440
| Metric | Agnes-3.0-Flash Preview | Inkling | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Agnes AI | Thinking Machines | — |
| Family | Other | Other | — |
| License | Apache 2.0 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Sep 13, 2026 | Jul 15, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | OpenRouter / Tinker (ref.) | — |
| Modalities | text, image, video → text | text, image, audio → text | — |
| Specs | |||
| Context window | 262K | 1.0M | B +786K |
| Max output | - | 131K | — |
| Parameters | 33B | 975B (41B act.) | B +942 |
| Pricing | |||
| Input $/1M | — | $1 | — |
| Output $/1M | — | $4.05 | — |
| Blended $/1M (3∶1) | - | $1.76 | — |
| Speed | |||
| tok/s | - | 55 | — |
| TTFT (s) | - | 0.55 | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 85.0 | 87.2 | B +2.2 pts |
| Humanity's Last Exam | — | 31.9 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 23.0 | 41.6 | B +18.6 pts |
| AA-LCR v1.1 | 68.3 | 77.3 | B +9.0 pts |
| CritPt | — | 5.4 | — |
| MMMU-Pro | — | 73.5 | — |
| IFBench | 74.2 | — | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | 77.6 | — |
| SWE-bench Pro | — | 54.3 | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 55.1 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 38.1 | 47.0 | B +8.9 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,410 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,440 | — |
Agnes-3.0-Flash Preview
- Context
- 262K
- Parameters
- 33B
- Price
- —
- Speed
- —
- Modalities
- text, image, video → text
- License
- Apache 2.0
Agnes AI's (Sapiens AI, Singapore) first open-weight release: a 33B-parameter hybrid-attention model (54 gated delta-rule recurrent layers alternating with 18 global-attention layers, 3:1) under Apache 2.0, with a 262K-token context window and text/image/video input. This open-weight Preview checkpoint is distinct from Agnes AI's production "Agnes 3.0 Flash" API model, which runs a different checkpoint and is not open-weight; the two should not be conflated.
Inkling
- Context
- 1.0M / 131K out
- Parameters
- 975B (41B act.)
- Price
- $1 / $4.05
- Speed
- 55 tok/s · 0.55s TTFT
- Modalities
- text, image, audio → text
- License
- Apache 2.0
Thinking Machines' first open-weights MoE (975B / 41B active) — natively multimodal with variable thinking effort and 1M context; Apache 2.0 weights on Hugging Face.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.