Vercel adds Kimi K3.1 Pro and Pro Highspeed to AI Gateway
Moonshot lists Pro cheaper than the earlier K3 and says Highspeed runs 2-3x faster than Pro at twice its price; Vercel offers both through AI Gateway.
By Benchdict AI DeskPublished Oct 23, 20263 min read
Vercel’s October 23 announcement says Moonshot AI’s Kimi K3.1 Pro and Pro Highspeed are available through AI Gateway, its interface for accessing AI models. On its own API platform, Moonshot lists Pro at $2/$10 per million input/output tokens and the faster Highspeed tier at $4/$20.
Key facts
- Pro input
- $2 per million tokens on Kimi API
- Highspeed input
- $4 per million tokens on Kimi API
- Highspeed speed claim
- 2-3x faster than Pro, Moonshot says
- Context window
- 1 million tokens for both models
- Model weights
- Release planned by November 4
Prices across the Kimi lineup
Moonshot says Highspeed runs two to three times as fast as Pro. Its API platform lists Highspeed at twice Pro’s input and output prices, while Pro is cheaper than the previous-generation K3.
| Model | Input per million tokens | Output per million tokens |
|---|---|---|
| Kimi K3 | $3 | $15 |
| Kimi K3.1 Pro | $2 | $10 |
| Kimi K3.1 Pro Highspeed | $4 | $20 |
The Kimi API platform lists cache writes—the charge for storing tokens for later reuse—at $2 per million tokens for Pro and $4 for Highspeed. Cache hits, the reuse of stored tokens, cost $0.20 per million tokens for either.
Moonshot describes Highspeed as retaining Pro’s intelligence and one-million-token context window, the amount of material a model can hold in a request. Pro also retains image input.
What Gateway access changes
Vercel’s announcement lists access through its AI SDK, a software development kit, and through the Chat Completions, Responses and Anthropic Messages API formats. The model identifiers are moonshotai/kimi-k3.1-pro and moonshotai/kimi-k3.1-pro-highspeed.
For users of coding agents such as Claude Code, Codex and Cursor, the announcement directs developers to install Vercel’s latest command-line tool and run its AI Gateway setup command.
Moonshot also lists availability through Kimi.ai, Kimi Work, Kimi Code and its own API.
Vercel says Gateway passes through provider pricing without markup and charges no platform fee on inference, including requests using a customer’s own provider key. Its separate custom-reporting announcement lists charges for tag or user-ID values and reporting queries.
The update from K3
Moonshot describes K3.1 Pro as the first model in the K3.1 series and a post-trained version of Kimi K3. On Moonshot’s own internal long-horizon coding benchmark, the reported score rose from 55.95% for K3 to 61.2% for Pro.
On Terminal-Bench 4.0, a benchmark Moonshot selected for comparison, the company reports 65.7% for Pro, against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5.
For developers seeking downloadable models, Moonshot plans to release the full weights—the files needed to run the model—by November 4.
Not yet confirmed
- The Vercel announcement was not directly accessed; Gateway availability and integration details are attributed to that announcement. Independent reporting has not corroborated the launch details here.
- Model-specific AI Gateway rates, request quotas, concurrency caps and maximum output-token limits are unconfirmed. The numerical prices above are Kimi API platform rates.
- Highspeed’s absolute latency, throughput and speed-comparison conditions are unconfirmed. Moonshot’s API platform uses 'roughly three times,' differing from its announcement’s two-to-three-times range; independent reproduction of the benchmark results is also unconfirmed.
- The planned weight release leaves the license, commercial-use terms and serving-hardware requirements unconfirmed.
What to watch
- Whether Moonshot releases the full model weights by November 4, with licensing and deployment requirements.
- Model-specific Gateway pricing and production measurements comparing Pro with Highspeed.
Sources
- Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available now in AI GatewayVercel Blog, Oct 23, 2026.
- Kimi K3.1 Pro and K3.1 Pro HighspeedHacker News (front page), Oct 23, 2026.
- Kimi API PlatformKimi API Platform, Oct 23, 2026.
- Custom reporting now available on AI Gateway - VercelVercel, Oct 23, 2026.
Benchdict AI Desk, AI-assisted reporting desk. Stories follow our editorial standards; corrections are logged above and on the corrections page.