Vercel adds DeepSeek V4.1 Flash 1017 to AI Gateway
Vercel reports faster tasks and shorter outputs. External measurements cover the Max variant, while Gateway-specific rates and configuration remain unconfirmed.
By Benchdict AI DeskPublished Oct 21, 2026Updated Oct 22, 20264 min read
Vercel says DeepSeek V4.1 Flash 1017 is available on AI Gateway, adding an update it describes as improving coding, reasoning and tool-assisted visual tasks. Its October 17 announcement promises provider pricing without an inference markup but supplies no model-specific rate table.
Key facts
- Gateway model identifier
- deepseek/deepseek-v4.1-flash-1017
- Reported changes
- Vercel says tasks finish faster and outputs are shorter than with the previous snapshot.
- External evaluation
- Artificial Analysis lists an Intelligence Index score of 45 for the Max variant.
- Listed Max pricing
- Artificial Analysis lists $0.30 per million input tokens and $1.20 per million output tokens; Gateway rates are unconfirmed.
A coding update with two access paths
Vercel serves the model through AI Gateway, its unified API for calling models. It describes 1017 as a minor update, with coding improvements covering terminal work and whole-repository building and editing. Its claim of better chart reading and visual reasoning specifically depends on tool access; it is not an unqualified claim about image understanding.
For developers already calling DeepSeek directly, the change may require no new model selection. Eesel and SmartScope report that the existing deepseek-flash alias automatically serves 1017. Gateway instead exposes the dated identifier deepseek/deepseek-v4.1-flash-1017.
For application maintenance, the direct-service reports describe a change behind an existing alias, while Vercel gives Gateway users a specific model name to request. Neither establishes access to the previous snapshot or a rollback path.
What the performance figures establish
Eesel reports that DeepSeek's own comparison table puts 1017 at 42.3 on Terminal-Bench 4.0, against 31.2 for the September build. It identifies those launch comparisons as vendor runs, not independent replications.
There is a separate external measurement. Artificial Analysis reports an Intelligence Index score of 45 and output speed of 216.5 tokens per second for the Max variant. Its page does not say what Max refers to; DeepSeek's own API offers reasoning-effort levels of low, high and max. It also lists text and image input, text output and a 1-million-token context window.
Those results provide a reference point for the model, but they do not verify Vercel's claimed improvement over the previous snapshot. The mapping between Artificial Analysis's Max configuration and the Gateway endpoint remains unconfirmed, as do Gateway-specific limits. No comparable rival-model results are established here, so these figures do not support a competitive ranking.
Listed token prices are not a Gateway quote
The available prices describe different service conditions:
| Service or listing | Input per million tokens | Output per million tokens |
|---|---|---|
| Artificial Analysis, Max variant | $0.30 | $1.20 |
| DeepSeek direct service, off-peak, reported by Eesel | $0.15 | $0.60 |
Eesel says the direct service's off-peak prices are unchanged. Neither row establishes what a particular Gateway provider route will charge, including cache treatment or time-of-day conditions.
Vercel says Gateway passes through provider prices without markup or an inference platform fee, including bring-your-own-key requests. That is a billing policy, not confirmation that either listed rate applies to this endpoint. Likewise, shorter outputs could reduce billed output volume, but Vercel supplies no workload-specific measurements establishing actual savings.
Who is affected
Coding-agent users get another access route: Vercel directs Claude Code, Codex and Cursor users to its latest CLI and AI Gateway setup command. Application teams also get Gateway's unified API, usage and cost tracking, retries and failover configuration, according to Vercel.
For teams managing shared spending, Gateway's API-key documentation permits optional caps. Requests using member-attributed keys are rejected when either the member budget or the team budget is exceeded—a separate operational constraint from the model's advertised capabilities.
Not yet confirmed
- Gateway-specific provider rates, caching treatment and time-of-day billing conditions are not confirmed.
- The Gateway endpoint's reasoning configuration and its correspondence to Artificial Analysis's Max variant are not confirmed; neither are Gateway-specific context, output, request-rate or concurrency limits.
- Production runtime and output-token savings, and independent replication of DeepSeek's before-and-after gains, are not established.
- The exact Gateway activation time, regional rollout and access to the previous snapshot are not confirmed.
- Model-specific retention, security and compliance conditions are not established.
- The Hugging Face model card yielded too little readable text to verify deployment specifications; self-hosting requirements and licensing are not established here.
What to watch
- A model-specific Gateway rate table and endpoint limits.
- Independent before-and-after evaluations using matched reasoning settings.
- First-party documentation on snapshot pinning and rollback.
Sources
- DeepSeek V4.1 Flash 1017 now available on AI GatewayVercel Blog, Oct 17, 2026.
- DeepSeek V4.1 Flash 1017 (Max)Hacker News (front page), Oct 19, 2026.
- AI Gateway API Keysvercel.com, Oct 20, 2026.
- DeepSeek V4.1 Flash 1017: what changed, benchmarks, and real costeesel.ai, Oct 18, 2026.
- DeepSeek-V4.1-Flash-1017: Changes, Benchmark Comparisons, API Pricing, and SpecificationsSmartScope, Oct 18, 2026.
- Thinking modeDeepSeek API Docs, Oct 22, 2026.
Benchdict AI Desk, AI-assisted reporting desk. Stories follow our editorial standards; corrections are logged above and on the corrections page.