vs
4:2
overall capability wins (0 ties across 6 scored). Agnes-3.0-Flash Preview leads overall.
- Reasoning4:0
- Specs0:2
Verdict
Agnes-3.0-Flash Preview edges reasoning & knowledge; Seed-OSS-36B-Instruct offers ~2.0x larger context (512K vs 262K).
- Agnes-3.0-Flash Preview edges reasoning & knowledge
- Seed-OSS-36B-Instruct offers ~2.0x larger context (512K vs 262K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Agnes-3.0-Flash Preview
- -
- Seed-OSS-36B-Instruct
- $0.35
Price per 1M tokens
Full comparison
Organization
Agnes AI
ByteDance Seed
Family
Other
Other
License
Apache 2.0
Apache 2.0
Open weights
Yes
Yes
Release
Sep 13, 2026
Aug 20, 2025
Knowledge cutoff
-
-
API / provider
No primary API price listed
SiliconFlow
Modalities
text, image, video → text
text → text
Specs
Context window
262K
512K
Max output
-
-
Parameters
33B
36B
Pricing
Input $/1M
—
$0.21
Output $/1M
—
$0.57
Blended $/1M (3∶1)
-
$0.30
Speed
tok/s
-
-
TTFT (s)
-
-
Reasoning
MMLU-Pro
—
—
GPQA Diamond
85.0
72.6
Humanity's Last Exam
—
9.9
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
23.0
18.2
AA-LCR v1.1
68.3
61.3
CritPt
—
—
MMMU-Pro
—
—
IFBench
74.2
41.9
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
—
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
38.1
—
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
—
| Metric | Agnes-3.0-Flash Preview | Seed-OSS-36B-Instruct | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Agnes AI | ByteDance Seed | — |
| Family | Other | Other | — |
| License | Apache 2.0 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Sep 13, 2026 | Aug 20, 2025 | — |
| Knowledge cutoff | - | - | — |
| API / provider | No primary API price listed | SiliconFlow | — |
| Modalities | text, image, video → text | text → text | — |
| Specs | |||
| Context window | 262K | 512K | B +250K |
| Max output | - | - | — |
| Parameters | 33B | 36B | B +3.00 |
| Pricing | |||
| Input $/1M | — | $0.21 | — |
| Output $/1M | — | $0.57 | — |
| Blended $/1M (3∶1) | - | $0.30 | — |
| Speed | |||
| tok/s | - | - | — |
| TTFT (s) | - | - | — |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 85.0 | 72.6 | A +12.5 pts |
| Humanity's Last Exam | — | 9.9 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 23.0 | 18.2 | A +4.8 pts |
| AA-LCR v1.1 | 68.3 | 61.3 | A +7.0 pts |
| CritPt | — | — | — |
| MMMU-Pro | — | — | — |
| IFBench | 74.2 | 41.9 | A +32.3 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | — | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 38.1 | — | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | — | — |
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.
Seed-OSS-36B-Instruct
- Context
- 512K
- Parameters
- 36B
- Price
- $0.21 / $0.57
- Speed
- —
- Modalities
- text → text
- License
- Apache 2.0
ByteDance Seed's open 36B dense model with a 512K native context window and a user-controllable reasoning 'thinking budget'.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.