Kimi K3 Now Available via Telnyx Inference API

(telnyx.com)

69 points | by fionaattelnyx 9 hours ago

16 comments

  • zius 11 minutes ago
    In my first interaction ("hi there kimi k3!"), Kimi K3 identified twice out of three times as Claude:

    > Hi there! Quick note — I'm actually Claude, made by Anthropic, not Kimi. But no worries!

    https://imgur.com/a/jqpc2Jc

    and

    > Just a quick heads-up — I'm Claude, made by Anthropic, not Kimi! Kimi is a different AI assistant (made by Moonshot AI), so it looks like there might be a little mix-up.

    https://imgur.com/a/AKxeysH

  • Mossy9 1 hour ago
    Also available from Nebius, via Cortecs: https://cortecs.ai/detailedServerlessView/kimi-k3

    €2.693/M input €13.464/M output Surprisingly, cache is not mentioned

    Upd: Tensorix joined the fray, with the same prices, with cache at €0.673/M read

  • madhu_ghalame 36 minutes ago
    Consider publishing latency, throughput, and cost metrics under different workloads to help teams make informed decisions.
  • rvz 13 minutes ago
    Jevon's paradox depends on how cheap the tokens get as the price of tokens get driven to zero as intelligence gets better and cheaper.
  • theredsix 5 hours ago
    10% cheaper than official! Let the inference pricing wars begin!
  • bedros 42 minutes ago
    any of these providers are HIPAA compliant?
  • smallerize 9 hours ago
    Very cool. What are your throughput and latency like?
  • TZubiri 27 minutes ago
    I think telnyx is a good product, with the only stain to its name being the supply chain attack on their python library.

    But I don't feel like providing inference is a professional move, it feels like out of scope for a telephony IaaS company. Feels like a FOMO moment where a reputation of years is crashed in a couple of weekends of being drawn into a fad.

    And the fact that it's a chinese model doesn't quite help? I guess it's on brand with the 'cheap' pay as you go brand telnyx might already be associated to.

    But more so it reads like Telnyx is trying to 'jump' into the trend of the 'open weights' discussion to compete with closed source incumbents. But we are at the tail end of the boom, anti ai sentiment is ever growing, customers now despise AI, especially in support channels, which is presumably the hook that Telnyx would have into 'AI'(LLMs). At this stage any company or individual that tries to join into the buzzword fueled cycle will pay the full fixed cost reputational price, but only reap the leftover hay from when the sun shone.

    AI(LLM) on support channels is essentially a decapitalization of a company/brand, the company has a reputation that customers value, and might be worth good money in the market, and by implementing AI (LLMs) on support, a lot of costs can be cut, while the brand loses value, not sustainable. And by Telnyx (or any B2B company)asking their clients to participate in this decapitalization move, they essentially gamble their reputation as well, albeit with better odds as shovel sellers.

  • jakswa 6 hours ago
    what quantization? FP4?
    • NitpickLawyer 2 hours ago
      The model is native mxfp4 w/ mxfp8 activations via QAT.
  • buffer_overlord 9 hours ago
    That’s huge
  • LoganDark 6 hours ago
    Telnyx was cool until they started demanding KYC. I would use them for burner phone numbers until they started saying they needed my government ID. Fuck that.

    OVH same thing. Tried to buy a VPS from them some years back and they said no VPS unless I provided ID. Would not refund me. Tried to dispute but my bank just gave me a credit instead.

    • illliillll 1 hour ago
      You can go on telegram and pay someone $10 to do the KYC for you.
    • TZubiri 23 minutes ago
      Seems like a good riddance, telnyx is for business usecases, not for personal use.

      Such use would be incompatible because it would lure in fraud, which would ruin the reputation of shared comms resources like ip blocks, phone number blocks, etc..

      If you are doing real business, you are providing KYC as a daily matter in procurement, thereby protecting consumers from Sybil scum.

    • nujabe 3 hours ago
      There were new regulations passed to combat robocalls that forced these companies to tighten up.
  • nttylock 58 minutes ago
    [flagged]
  • marsven_422 3 hours ago
    [dead]
  • hiherer 5 hours ago
    [dead]
  • teravor 5 hours ago

        > Because we own the infra, the per-token price reflects the cost of running the model, not the cost of renting someone else's plus their margin.
    
    
    is not compatible with

        > Pricing on Telnyx: $2.70/1M input tokens, $13.50/1M output tokens, $0.27/1M cached input tokens.
    
    since you asserted something false and bizarre, how about telling us what is your markup?
    • alexeldeib 5 hours ago
      Why are those incompatible? Pricing is 10% under moonshot, and pure infra providers also want money
    • gruez 5 hours ago
      I don't get it, what's the contradiction supposed to be?
    • rubslopes 5 hours ago
      They are selling it even cheaper than Moonshot AI. Why are you so sure they are lying?
      • teravor 4 hours ago

            We estimate that the true blended price per million tokens for running Opus 4.7 on agentic tasks at $0.99 despite the sticker price being $5/$25 per MTok.
        
        https://newsletter.semianalysis.com/p/ai-value-capture-the-s...

        according to Semianalysis those prices would be far above actual costs.

        • petu 38 minutes ago
          They're not saying anything about Anthropic serving costs in that quote, just calculating what MTok price is for running agents. Next sentence after your quote:

          > Agentic workloads have extremely high input-to-output ratios (our Claude Code usage has a ratio of about 300:1) and high cache hit rates (90%+). Because cached input tokens only cost $0.50/MTok, most of the tokens end up in the cheapest tier.

          90% cache hit input blend: 0.9 * $0.5 + 0.1 * $5 = $0.95 per MTok.

          300:1 input/output blend: (300 * $0.95 + $25) / 301 = $1.03 per MTok.

          They don't say exact cache hit rate they calculated for ("90%+"), so close enough.

        • tpm 46 minutes ago
          If they are indeed far above actual costs, then surely price discovery will be done by the overall market in due course.
        • FergusArgyll 4 hours ago
          **reflects** the cost...

          Not the **literal** cost