Low-latency multimodal reasoning model with a 262K window, taking text, image and video input for high-volume pipelines.
Modality:
Text
Image
model ID
bytedance-seed/seed-1.6-flash
Output Speed *
N/A
tok/s
Intelligence Index *
N/A
/ 100
Context Window *
262144
tokens
Input price
0.45
Anytoken
Output price
1.8
Anytoken
Seed 1.6 Flash: Fast, Low-Cost Multimodal Reasoning at 262K Context
Seed 1.6 Flash is the speed-oriented tier of ByteDance Seed's Seed 1.6 family. It is a multimodal reasoning model that accepts text, image and video input and returns text. Its route metadata lists a 262,144-token context window, tool calling, structured outputs and reasoning parameters. Independent gateway statistics show sub-second median latency and moderate output speed. It suits high-volume classification, media understanding and agent loops where per-request cost matters more than peak reasoning quality.
Test Seed 1.6 Flash on your own multimodal prompts via AnyAPI.ai.
Performance
Sub-Second Latency, Moderate Throughput, Unproven Reasoning Depth
Seed 1.6 Flash is the lightweight end of ByteDance's Seed 1.6 line. It combines a 262,144-token window with text, image and video input. OpenRouter's public statistics show a median latency of 0.71 seconds and about 52 output tokens per second. That is responsive for interactive use but not a record throughput figure, so "ultra-fast" is best read as fast for its tier. It fits high-volume routing, classification and multimodal triage. No strong independent reasoning or coding scores were verified, so test quality on your own prompts.
Benchmarks
Seed 1.6 Flash Speed and Task Evidence from Independent Tests
Independent evidence is limited to operational and task-level tests. OpenRouter's route statistics list about 52 tokens/s and 0.71 s median latency. Respan's gateway reports 64 tokens/s and about 1.0 s latency over 30 days. In AiPy's agent-task evaluation, Flash had the shortest average completion time at 73 seconds, though the report says it traded off accuracy. No standard intelligence or coding leaderboard scores for this exact model were verified.
Output Speed
*
N/A
tok/s
Intelligence Index
*
N/A
/ 100
MMLU *
Broad world knowledge and problem-solving
0
%
GPQA *
PhD-level scientific reasoning across physics, biology, chemistry.
0
%
HLE *
Adherence to multi-step structured instructions.
0
%
LiveCodeBench *
Tool-calling reliability in long agentic loops.
0
%
Technical Specifications
What the model supports
Seed 1.6 Flash accepts text, image and video and returns text only. Its context window is 262,144 tokens. The output ceiling differs by source: the provider description cites 16K tokens, while OpenRouter route metadata lists 32,768. Keep the two limits separate. The large window suits long inputs, but long generations should be capped conservatively and tested. Reasoning, tool and structured-output parameters are listed, but specific reasoning effort levels were not verified for this model.
Verified Specifications —
Seed 1.6 Flash
*
Input modalities
Text
Image
output modalities
Text
Context window
262144
tokens
Maximum output tokens
32768
Reasoning
Yes
Knowledge cutoff
June 2025
Pricing (standard)
0.45
AnyTokens in
/
1.8
AnyTokens out
Comparison
Seed 1.6 Flash or Seed 2.0 Mini for Low-Cost Multimodal Workloads?
Seed 2.0 Mini is ByteDance's lightweight tier of the newer Seed 2.0 family. Third-party listings describe it as released in February 2026, with a 256K context window, text, image and video input, and four reasoning effort levels (minimal, low, medium, high). Like Seed 1.6 Flash, it targets low-latency, cost-sensitive traffic. The practical decision is whether to stay on a known Flash-tier model or move to its newer lightweight successor, especially since Flash is listed for removal on OpenRouter.
Dimension
Seed 1.6 Flash
Seed-2.0-Mini
Context window *
262144
tokens
262144
tokens
Output speed *
N/A
tok/s
N/A
tok/s
Intelligence Index *
N/A
5.5
Input pricing
0.45
AnyToken
0.6
AnyToken
Output pricing
1.8
AnyToken
2.4
AnyToken
Knowledge cutoff *
June 2025
February 2026
Choose Seed 1.6 Flash when you have already validated it on your prompts, want the lowest-cost lightweight tier, and do not need explicit reasoning effort levels. Choose Seed 2.0 Mini when you want controllable reasoning effort, a newer generation with a longer expected availability window, or a migration target before Flash is retired. Seed 2.0 Mini is also the safer choice for new builds. No head-to-head independent benchmark for the two was verified.
Limitations & Trade-offs
1
Scheduled removal risk. OpenRouter's Seed provider page lists Seed 1.6 Flash as going away on November 11, 2026, and Seed 1.6 carries the same notice. This affects any long-lived deployment built on a fixed model ID. Confirm current availability before committing, and plan a migration path to a newer ByteDance tier such as Seed 2.0 Mini.
2
Reasoning and coding quality are unproven. No standard intelligence or coding leaderboard scores for this exact model were verified. A three-class sentiment study reported overall accuracy of 0.69 for Doubao-seed-1.6-Flash, which it called moderate, and AiPy noted an accuracy trade-off for its speed. Use Flash for first-pass work, and escalate complex reasoning, code agents or high-precision extraction to a stronger model.
3
Text-only output and an ambiguous output cap. The model returns text only, with no image, audio or video generation. Sources also disagree on the output ceiling: the provider description cites 16K tokens, while OpenRouter metadata lists 32,768. Workloads that depend on very long single generations, such as full-document rewrites, should set conservative limits and verify truncation behavior before launch.
4
Reasoning fields can break streaming clients. A GitHub issue reports that Seed models, including Seed 1.6 Flash via OpenRouter, return reasoning and reasoning_details fields in streaming deltas that opencode's parser does not handle, causing hangs. Teams using custom streaming parsers or agent harnesses should handle these fields explicitly. Measured output of about 52 tokens/s is also moderate, not the fastest available.
Best-Fit Workloads
Where this model earns its place
01
Image and video triage
Seed 1.6 Flash accepts image and video input alongside text and returns text, which fits tagging, captioning, content pre-screening and asset classification at volume. Its lightweight tier and sub-second median latency on OpenRouter suit large queues. Output is text only, and no independent multimodal benchmark was verified. Validate label accuracy on a sample and route ambiguous items to a stronger vision model.02
First-pass classification and routing
As a Flash-tier model, Seed 1.6 Flash fits intent detection, ticket triage and content labeling, where each request is short and volume is high. Structured output parameters help constrain results to a schema. Independent accuracy evidence is moderate (0.69 in one sentiment study), so treat it as a first pass and send low-confidence cases to a higher-accuracy model.03
Latency-sensitive agent loops
Tool calling, tool_choice and reasoning parameters are listed, and AiPy recorded Flash as the fastest model in its agent-task test at 73 seconds average. That suits quick tool-selection and lightweight multi-step workflows. AiPy also noted an accuracy trade-off, so use Flash for fast sub-steps and a stronger model for planning or final verification.04
Long-document extraction
The 262,144-token window lets Seed 1.6 Flash take long contracts, logs or transcripts in one request, and structured outputs can return fields as JSON. This suits bulk extraction where cost per document matters. Cap output length and spot-check completeness. For nuanced multi-document reasoning, a higher-tier Seed model is the safer choice.Integration
01
One API integration
02
Easy model switching
03
Flexible for production
04