Philip Skeete

@AskVenice I'm running @NanoClaw_AI via @AskVenice and GLM-5.1 is doing a remarkable job carrying the Claude Code SDK and Agent SDK harnesses. Mapping: Opus --> claude-opus-4-6, Sonnet --> zai-org-glm-5-1, Haiku --> zai-org-glm-4.6. This has been a super solid setup 👍

Apr 8

Spent today running a proper latency A/B on GLM-5.3-Flash across our two stake-for-inference providers: @AskVenice and @near_ai. Both let you stake tokens once and draw inference against that stake instead of paying per token. NEAR's leg runs inside a confidential TEE enclave with a signed attestation report you can actually pull and verify. Same 20-request probe set, same payload sizes, run back to back. Time to first token (median, small requests) • NEAR: 0.74s • Venice: ~1.1s Time to firs

Sep 2

Spent today running a proper latency A/B on GLM-5.3-Flash across our two stake-for-inference providers: @AskVenice and @near_ai. Both let you stake tokens once and draw inference against that stake instead of paying per token. NEAR's leg runs inside a confidential TEE enclave with a signed attestation report you can actually pull and verify. Same 20-request probe set, same payload sizes, run back to back. Time to first token (median, small requests) • NEAR: 0.74s • Venice: ~1.1s Time to firs

Sep 2

@ErikVoorhees Venice deploys a new model before I even know it exists, this is an un-paralleled rate of iteration 🤯

Apr 7

@ErikVoorhees @Wayland_Six @NousResearch This would be a great add to Venice 👍

Apr 17

@ErikVoorhees @burstingbagel Great post, $DIEM took me a long time to fully conceptualize, so I imagine it's the same for others. But once you get it, you get it (literally) 😀

May 6

@jietang I've been running GLM-5.1 as my Sonnet-4.6 replacement in @NanoClaw_AI via @AskVenice and I have to say it's works beautifully. My anecdotal take is Opus finds fewer errors with GLM-5.1 code than Sonnet-4.6. And GLM models carry the Claude Code and Agent SDK harnesses very well

Apr 11

@AskVenice @near_ai I've been chatting for hours and our agent's harness carries a base prompt of ~60K tokens, here's the bill: z-ai/glm-5.3-flash - $0.18, each turn is ~$0.0012 😂

Sep 2

Holy smokes, has anyone tried running their @openclaw or @NanoClaw_AI harness on @AskVenice's private Grok-4.20-beta model, I am getting near instantaneous responses, and I don't need a skill for X-search. I've only just started testing but so this a game changer, near Opus level reasoning, 10x faster and fully private at a fraction of the cost🤯 One caveat: tool use is a little sketchy, specifically file edit. GLM models handle the Claude Code SDK just fine, but Grok needs some training.

Apr 9

@nikshepsvn Good list, your first three are what I would call best current plays in crypto, in that order. I think $NEAR pulls from $ZEC, similar mind share, better fundamentals. Hard to argue with your $VVV thesis for top spot. $HYPE a strong asset but very crowded trade rn

Jun 19