Novita AI LLM

Hy4 Flash on Novita AI: Vision Support at 1/6 the Cost of Hy4

Hy4 Flash on Novita AI: Vision Support at 1/6 the Cost of Hy4

Hy4 Flash is now available on Novita AI’s OpenAI-compatible serverless API under the model ID tencent/hy4-flash. It lists $0.07 per 1M input tokens, $0.281 per 1M output tokens, $0.003 per 1M cached reads, a 977K context window, a 63K maximum output, and text and image input. Its input rate is about one-sixth of Hy4’s current $0.417, with the same context ceiling and the vision support that Preview lacks. Test Flash first for cost-sensitive agentic, long-context, and screenshot-heavy work; keep Hy4 where top-end reasoning matters.

Key Takeaways

  • Live since November 16, 2026. Hy4 Flash is a separate serverless model from Hy4 and Hy4 Preview; name the ID you intend to use.
  • Input costs about one-sixth of Hy4. $0.07 versus $0.417 per 1M input tokens; output is $0.281 versus $1.2505, about one-quarter.
  • Limits match Hy4. 977K context and 63K maximum output.
  • Vision and the same access surface. Text and image input, text output, and chat/completions, responses, and Anthropic-compatible endpoints with function calling, structured outputs, and reasoning.

Hy4 Flash Specs and Pricing on Novita AI

Novita AI lists Hy4 Flash as a serverless chat model.

FieldDetails
Display nameHy4 Flash
Model IDtencent/hy4-flash
Base URLhttps://api.novita.ai/openai
Endpoint familychat/completions, responses, Anthropic-compatible
ModalitiesText and image input, text output
Context window1,000,000 tokens listed; the model page displays 977K
Max output64,000 tokens listed; the model page displays 63K
Pricing$0.07 input, $0.281 output, $0.003 cached input per 1M tokens
Best fitCost-sensitive agentic, long-context, and image-assisted work that does not need Hy4’s top-end reasoning

Specs and pricing from the Novita AI model page, checked November 17, 2026. Plan budgets against the 977K and 63K figures.

What Is Hy4 Flash?

Hy4 Flash is the newest and least expensive member of the Hy4 line on Novita AI. Novita lists the same 1M-token context, 64,000-token output ceiling, endpoints, and features as Hy4, at a fraction of the price. The listing claims gains over Hy4 Preview on agentic and long-horizon tasks, plus visual understanding; test those claims yourself.

The concrete differences are in the listings. Hy4 and Hy4 Flash both accept images; Hy4 Preview is text-only. Flash is the cheapest of the three. Novita publishes no parameter count for Flash, so do not assume Preview’s 770B/49B architecture.

Hy4 Flash vs Hy4 and Hy4 Preview

ModelInput priceOutput priceContextImage input
Hy4 Flash$0.07$0.281977KYes
Hy4$0.417$1.2505977KYes
Hy4 Preview$0.834$2.501977KNo

Prices are per 1M tokens. Hy4’s $0.417/$1.2505 are current rates; its origin rates are $0.834/$2.501, so the gap widens if the lower rate ends. Checked November 17, 2026.

The one-sixth figure is precise for input ($0.417 versus $0.07, about 5.96x). Output is about 4.5x cheaper and cached reads 7x: input-heavy requests approach the headline saving, while long answers see closer to one-quarter the output bill.

Hy4 Flash API Access on Novita AI

Hy4 Flash runs on Novita AI’s serverless API with no deployment step. Point an OpenAI-compatible client at https://api.novita.ai/openai, set the model field to tencent/hy4-flash, and use a Novita AI API key. Anthropic-style clients use https://api.novita.ai/anthropic; responses is also available.

The listing includes function calling, structured outputs, reasoning, and serverless delivery. No separate vision parameter is documented; send images via standard chat completions content parts with an image_url URL or base64 payload. Test one screenshot for quality and token cost first. Field syntax is in the Create Chat Completion reference. For the request shape on the earlier Hy model, see the Hy3 API quick start.

Key Capabilities for Developers

The 977K-token window fits specifications, repository state, and retrieved documents in one request, though long context still needs curation.

Function calling and structured outputs make Flash an execution tier for agent loops: the model chooses tools and emits arguments your application validates. Reasoning support suits harder planning steps, though the listing does not specify modes or budgets. Image input covers screenshots, diagrams, and document pages alongside the instruction.

Test whether Flash retains Hy4-class agentic behavior at the lower price: compare task success, valid tool calls, retries, and cost per task on one set.

When to Use Hy4 Flash

  • Input-heavy long-context jobs. Repository analysis, document review, and multi-source summarization gain most from the one-sixth input rate.
  • High-volume agentic steps. Triage, extraction, routing, and tool loops that do not need Hy4’s strongest reasoning.
  • Screenshot and diagram workflows. UI review, chart reading, and document-page extraction with image input at a low price.
  • Routing and evaluation tiers. Default to Flash and escalate to Hy4 only where evaluation shows a gap.

When Not to Use Hy4 Flash

  • Hardest reasoning and coding. If Hy4 measurably wins on your failures, price is not the deciding number.
  • Output-dominated generation. Long answers cost about one-quarter the Hy4 output rate, not one-sixth of total cost.
  • Workloads needing a long production record. Flash is newly listed; stage traffic and keep a fallback ID.
  • Audio or video understanding. The listing covers text and image only.
  • Untested latency floors. Measure time to first token, tool-cycle time, and retries before routing interactive traffic.

How Hy4 Flash Fits Your API Workflow

Flash shares the base URL, endpoints, and request shape with Hy4, so routing is a model-field change. Keep the model ID in configuration, send routine steps to Flash, escalate to Hy4 on failures or difficulty thresholds, and log which ID served each request. Cached input reads at $0.003 per 1M tokens reward repeated system prompts and tool schemas; re-run your evaluation after any price change.

Conclusion

Hy4 Flash brings Hy4’s context length, endpoints, and image support to Novita AI at about one-sixth of Hy4’s current input rate. It suits cost-sensitive agentic and long-context work and high-volume screenshot and document pipelines, but is not a proven replacement on the hardest reasoning tasks. Explore Hy4 Flash on Novita AI and compare it against Hy4 on your own task set; keep the model ID in configuration so you can route by measured quality.

FAQ

What is the model ID for Hy4 Flash on Novita AI?

The exact model ID is tencent/hy4-flash. Use it in the model field with the OpenAI-compatible base URL; the Hy4 Flash model page lists current limits and pricing.

How much does Hy4 Flash cost on Novita AI?

As of November 17, 2026, Hy4 Flash lists $0.07 per 1M input tokens, $0.281 per 1M output tokens, and $0.003 per 1M cached reads. Confirm rates on the model page before budgeting.

Is Hy4 Flash different from Hy4 and Hy4 Preview?

Yes, all three are separate IDs. Hy4 Flash accepts text and images at $0.07/$0.281 per 1M tokens, Hy4 also accepts images at $0.417/$1.2505, and Hy4 Preview is text-only at $0.834/$2.501. All list 977K context and 63K maximum output.

Does Hy4 Flash support image input on Novita AI?

Yes. Hy4 Flash lists text and image input and text output; no separate vision parameter is documented. Send images as standard chat completions content parts; test a representative image first.

What are the context window and maximum output for Hy4 Flash?

Hy4 Flash lists a 1,000,000-token context window and 64,000-token maximum output; the model page displays 977K and 63K. Plan against the lower figures.

Is Hy4 Flash a good fit for coding agents?

Yes, as an execution tier. Hy4 Flash supports function calling, structured outputs, and reasoning at a far lower input rate. Test it on tasks Hy4 passes and fails, and keep Hy4 as the escalation path.

Related Posts