NVIDIA K3 listing: first checked September 14, terms rechecked October 3
NVIDIA’s Kimi K3 model card and live API catalog list Kimi K3. The hosted API remains governed by NVIDIA’s separate API Trial Terms. Sections 1.2 and 1.4 say the trial API and its generated content are not for production; production use requires a separate NVIDIA or service-provider subscription.
The terms prohibit confidential, controlled, or sensitive input. Personal data is allowed only if the specific API service expressly permits it. NVIDIA says it collects user content and generated content to improve products and services, including AI models; service-specific disclosures can add retention terms. Treat this as a trial with data-use conditions, not a private coding route. NVIDIA may extend trial credits, but the terms do not promise a universal amount or entitlement. Account access, payment, credits, quota, expiry, rate limits, and K3-specific retention were not tested. The NIM guide keeps its August K2.6 setup as history; archive capture of the current terms remains archive-pending.
Kimi now has a new flagship route, a cheaper coding route, and several older access paths that should not be mixed together.
| |
Current Kimi Routes
| Route | Model identity | Billing | Best for |
|---|---|---|---|
| Kimi K3 API | kimi-k3 | Pay as you go | Latest Kimi flagship, 1M context, multimodal/agentic evals |
| Kimi K2.7 Code API | kimi-k2.7-code | Pay as you go | Reproducible coding API evaluation |
| Kimi K2.7 Code HighSpeed | kimi-k2.7-code-highspeed | Pay as you go at 2x base token prices | Latency-sensitive coding |
| Kimi Code membership | k3, kimi-for-coding, provider-managed backend IDs | Subscription credits and quotas | Official CLI and supported coding agents |
| NVIDIA NIM trial | moonshotai/kimi-k3 listing | Trial; account-specific details untested | Bounded internal compatibility evaluation; see trial limits above |
K3 is the newest Kimi flagship and the right route for latest-Kimi intent. K2.7 Code remains the cheaper routine coding model. K2.6 remains a compatibility and historical NVIDIA setup/comparison lane; K2.5 remains historical context.
API Pricing
Prices are per 1 million tokens.
| Model | Cache-hit input | Cache-miss input | Output | Context |
|---|---|---|---|---|
| K3 | $0.30 | $3.00 | $15.00 | 1M |
| K2.7 Code | $0.19 | $0.95 | $4.00 | 256K |
| K2.7 Code HighSpeed | $0.38 | $1.90 | $8.00 | 256K |
| K2.6 | $0.16 | $0.95 | $4.00 | 256K |
K3 is priced as a flagship escalation lane, not a budget lane. Moonshot has released the model weights and serving materials under the Kimi K3 License. Treat them as open weights, not as a generic open-source grant; review that license and your serving requirements before self-hosting.
K2.7 Code requires thinking mode in the documented API path. HighSpeed is the same model at higher token prices, with Kimi advertising approximately 180 tokens/s and up to 260 tokens/s for short contexts while warning that constrained resources can make performance fluctuate.
Kimi Code Membership
The current Kimi Code model configuration shows K3 and K2.7 Code across plan-dependent model IDs:
| |
Membership-backed third-party tools can use k3 or kimi-for-coding, depending on plan and context entitlement. Membership k3 supports low/high/max reasoning; Moderato exposes up to 256K context, while Allegretto and higher tiers expose up to 1M. That is convenient but less reproducible than pinning kimi-k3 or kimi-k2.7-code in the API, where K3 is currently always-on and max-only.
Kimi documents weekly quota refresh, a rolling five-hour frequency window, shared device/API-key quota, and a unified monthly credit pool across membership features. Check the signed-in console for the current balance and exact plan terms.
The VS Code path is transitional: current docs say new extension installs are limited to legacy Python CLI users while the TypeScript CLI integration is adapted. Other editors can connect through ACP.
Free and Promotional Routes
NVIDIA’s trial and data-use limits are summarized above. Check the NIM guide and current account terms before evaluating it; no universal free quota or account access was verified.
- That does not make K2.7 Code free.
- A no-payment-required trial can be labeled Free only while the live signup requires no payment.
- A discounted first month is Low-cost, not free.
- Checkout screenshots and old campaign rules are not evergreen entitlements.
- Kimi’s current Moonshot Together rules describe a referral draw with published odds. The 30- and 365-day credit tiers require a paid-subscription task chance, awarded credits expire December 31, and AIHackers has not run redemption or published a referral CTA. Use the neutral draw guide.
NVIDIA NIM access and K2.6 history now links the K3 listing. The August K2.6 setup remains an older example, not a tested K3 integration. Recheck signed-in access, service-specific input/data terms, and the exact provider model ID before an evaluation.
For a free coding IDE rather than Kimi specifically, use the Free Stack guide and verify its current hosted model roster.
Evidence and Quality
benchmark artifact
Kimi K3, K2.7, and Dated Coding Comparisons
| Model | Provider | Status | Context | Input price | Output price | Coding signal | Tool-use signal | Benchmark evidence | Speed | Verdict | Sources | Checked |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Kimi K3 | Moonshot AI |
active Kimi API/product flagship; Kimi K3 License weights. NVIDIA also lists K3 under trial terms as checked September 14; account access untested. | 1M | $3.00 / 1M $0.30 cache-hit / $3.00 cache-miss input / $15.00 output per 1M tokens | $15.00 / 1M | Moonshot reports strong max-reasoning launch-suite coding and agent results; AIHackers repo eval is not verified. | Kimi API and Kimi Code support K3; Kimi docs warn to preserve full assistant history and avoid mid-session model switching. |
| Artificial Analysis reports 62 output tokens/s and flags high verbosity; measure total output cost per accepted task. | Test as Kimi's newest 1M-context frontier-adjacent lane; keep K2.7 Code for cheaper routine Kimi coding until K3 passes local CAR tests. | NVIDIA Kimi K3 model card, Kimi K3 launch blog [archive], Kimi K3 quickstart [archive], Kimi K3 API pricing [archive], Kimi K3 weights and license [archive], Kimi current model list [archive], Kimi Code model configuration, Artificial Analysis: Kimi K3 [archive] | 2026-08-01 |
| Kimi K2.7 Code | Moonshot AI |
active Cheaper routine Kimi coding API lane; HighSpeed is the same model at higher token prices. | 256K | $0.95 / 1M $0.19 cache-hit / $0.95 cache-miss input / $4.00 output per 1M tokens; HighSpeed doubles those rates | $4.00 / 1M | Kimi K2.7 Code remains the lower-cost 256K coding lane after K3; independent normalized benchmarks are not imported. | OpenAI-compatible API; thinking mode required in the documented K2.7 Code quickstart. |
| HighSpeed model ID exists at a higher token price; latency not independently measured here. | Cheaper routine Kimi coding API lane when Kimi routing fits and 256K context is enough. | Kimi K2.7 Code quickstart [archive], Kimi K2.7 Code pricing [archive], Kimi Code K2.7 release notes [archive], SWE-bench, Berkeley Function Calling Leaderboard | 2026-06-28 |
| GLM-5.2 | Z.AI |
historical Historical July comparison; Z.AI now documents GLM-5.3. These scores and prices do not evaluate 5.3. Historical status does not imply API retirement. | 1M | $1.40 / 1M $1.40 input / $0.26 cached input / $4.40 output per 1M tokens | $4.40 / 1M | Z.AI reports 62.1 on SWE-Bench Pro and 81.0 on Terminal-Bench 2.1. | Supported-tool coding lane; BFCL score not imported. |
| Artificial Analysis flags higher output-token use; measure total cost per successful task. | Dated July value candidate; retain for pinned integrations and historical comparisons. Check the newer GLM revision separately before new routing decisions. | Z.AI GLM-5.3 successor documentation, Z.AI GLM-5.2 overview [archive], Z.AI pricing [archive], Artificial Analysis: GLM-5.2 article [archive], Artificial Analysis Intelligence Index v4.1, SWE-bench, Berkeley Function Calling Leaderboard | 2026-06-28 |
| Gemini 3 Flash |
active Current Gemini value lane where Gemini API or Vertex AI fits. | 1.05M input | $0.50 / 1M $0.50 input / $3.00 output per 1M tokens | $3.00 / 1M | not verified | Function calling and code execution supported. | not verified | Preview model positioned for lower latency; independent value not imported. | High-context value lane when Gemini API or Vertex AI fits. | Gemini API models, Gemini API pricing, Artificial Analysis: Gemini 3 Flash, LMArena leaderboard dataset | 2026-05-26 | |
| Claude Sonnet 5 | Anthropic |
active Generally available across Claude plans, Claude Code, the Claude API, GitHub Copilot, and supported AWS paths. | 1M | $2.00 / 1M $2.00 input / $10.00 output checked September 27; earlier launch schedule superseded | $10.00 / 1M | Anthropic reports substantial coding and agentic gains over Sonnet 4.6; independent normalized results are pending. | Available in Claude Code and the Claude API; adaptive thinking is on by default. |
| No site-owned normalized latency result is verified. | First Claude cost/performance test before Opus 5.5; escalate only when the premium pass changes the accepted result. | Claude Sonnet 5 current specifications, Anthropic Claude Sonnet 5 launch [archive], Claude Sonnet 5 migration guide [archive], GitHub Copilot Claude Sonnet 5 launch [archive], Claude Sonnet 5 on AWS [archive], Artificial Analysis: Claude Opus 5 [archive] | 2026-09-27 |
| Claude Opus 4.8 | Anthropic |
historical Historical comparison; use Opus 5.5 for current premium evaluation. | 1M | $5.00 / 1M $5.00 input / $25.00 output per 1M tokens | $25.00 / 1M | Historical premium Claude baseline; use Opus 5 for new task-level comparisons. | Still available for pinned integrations; new Claude premium routing should test Opus 5. |
| Artificial Analysis measured 57.3 output tokens/s; provider and workload latency vary. | Historical premium baseline. Use Claude Opus 5.5 for current Claude premium routing. | Claude models overview [archive], Claude API pricing [archive], Artificial Analysis: Claude Opus 4.8 [archive], Artificial Analysis Intelligence Index v4.1, LMArena leaderboard dataset, Berkeley Function Calling Leaderboard | 2026-07-25 |
K3 has early independent evidence but no AIHackers-owned repository result. K2.7 improvements remain vendor-reported.
Kimi K3 has an Artificial Analysis Index 57 row and Moonshot launch-suite evidence. Kimi’s earlier K2.7 release notes report improvements over K2.6 of 10.4% on Program-Bench, 11.4% on MCP Mark Verified, and 76.2% on SWE Marathon, plus 30% lower reasoning-token use. Keep the source labels attached.
Do not reuse K2.5’s 76.8% SWE-bench Verified result as a K2.7 score.
Cost per Successful Task
Run the same repository task and record:
- exact model ID or backend alias;
- input, output, cache hits, and cache writes;
- retries and wall-clock time;
- tests and accepted completion;
- human review and repair.
The HighSpeed premium is justified only if lower latency changes the workflow. The membership is justified only if its credits and tool convenience beat measured API use for the same accepted tasks.
Which Route to Choose
| Requirement | Recommended route |
|---|---|
| Newest Kimi flagship and 1M context | K3 API or K3-enabled Kimi Code plan |
| Explicit current model and repeatable API test | K2.7 Code API |
| Lowest K2.7 cache-hit cost | Base K2.7 Code |
| Lowest latency | HighSpeed after measuring the premium |
| Official Kimi CLI and third-party coding agents | Kimi Code membership |
| One-off zero-dollar test | A currently verified trial, with the actual model labeled |
| 1M context at lower token price | Compare GLM-5.2, Gemini, MiniMax, or Xiaomi instead |
| Premium final review | Claude Opus 4.8 or another approved premium lane |
Sources
- Kimi API: Global model and pricing surface
- Kimi: K3 launch blog (Archive)
- Kimi Platform: K3 quickstart (Archive)
- Kimi Platform: Current model list (Archive)
- Kimi Platform: K3 pricing (Archive)
- Kimi: K3 weights and license (Archive)
- Kimi: Kimi Code model configuration (fresh archive pending after exact/protocol/wildcard retries and an HTTP 520 save response on July 18)
- Kimi Platform: K2.7 Code quickstart (Archive)
- Kimi Platform: K2.7 Code pricing (Archive)
- Kimi: Kimi Code
- Kimi: Kimi Code documentation (Archive)
- Kimi: K2.7 Code release notes (Archive)
- Kimi: Membership credit update (Archive)
- Kimi: Moonshot Together rules — live rules rechecked August 20; archive
archive-pending - NVIDIA: Live API model catalog
- NVIDIA: Kimi K3 model card
- NVIDIA: API Trial Terms — rechecked 2026-10-03; archive
archive-pending - NVIDIA: Kimi K2.6 reference
Related links
- /models/kimi-k3/
- /models/kimi-k2.7-code/
- /tools/kimi-code/
- /compare/models/glm-5.2-vs-kimi-k2.6/
- /compare/codex-vs-claude-vs-kimi/
- /value/free-stack/
- /value/deals/kimi-referral-draw/
- /value/llm-cost-saving-playbook/
- /risks/kimi/
NVIDIA’s Kimi listing and API Trial Terms were rechecked October 3, 2026. Other Kimi access, pricing, membership, and promotional content retains its prior review dates; NVIDIA account access and K3-specific data handling remain untested.