AUGUST 5, 2026 — GROK VOICE TF 2.0 LIVE · WEEK ROUNDUP
- Grok Voice Think Fast 2.0 — now the default: grok-voice-latest switched today. AA agentic #1 (56.5%), STS #2 overall (82.9%). First audio: 0.70s. Price: $0.08/min (+60% from v1.0). Starlink A/B: higher sales conversion and support containment. Pin grok-voice-think-fast-1.0 to stay on v1.0. Qwen Audio 3.0 TTS Plus holds the STS #1 overall slot. Full review →
- Week in review — the stories that matter:
• Palantir Q2: $1.94B revenue +93%, US commercial +149%, Rule of 40 at 155%. AI enterprise revenue is now real at scale.
• Gemini 3.6 Flash: $1.50/M, faster (304 tokens/sec), 17% fewer output tokens — AA Intelligence Index stayed at 50. Migrate from 3.5 Flash. Evaluate vs Luna before migrating from anything else.
• Qwen3.8-Max: previewed July 19, not an August launch. No benchmark table. No confirmed API price. No open weights. Test via Token Plan.
• August 31 triple deadline: Kimi kimi-k2.5 sunset · Claude Sonnet 5 $2→$3/M · Both on the same day. Act now.
Story 1 — Grok Voice Think Fast 2.0: The Migration Is Complete
As of August 5, 2026, grok-voice-latest is now Think Fast 2.0 — per SpaceXAI's official announcement. Any team that did not pin grok-voice-think-fast-1.0 before today is now on 2.0 at $0.08/min. The AA benchmark shows Think Fast 2.0 leading on the agentic score (56.5% vs GPT-Realtime-2.1's 45.7%) and the overall STS Quality Index (82.9% vs 79.1%). The Starlink production A/B test — higher sales conversion and support containment — is the real-world evidence that the benchmark gains translate to business outcomes. To roll back: pin grok-voice-think-fast-1.0. To verify your current model: pull a usage export and confirm the model ID in your billing. Full review and benchmark breakdown →
Week Roundup — August 1-5, 2026
Palantir Q2 2026 settled the debate: AI enterprise revenue is real. $1.94 billion (+93% YoY), US commercial +149%, 220 deals of $1M+ in a single quarter, Rule of 40 at 155%. Karp's "AI sovereignty" framing is not just positioning — it is closing contracts at scale. Full analysis →
Gemini 3.6 Flash (launched July 21) reviewed: $1.50/$7.50/M, 304 tokens/sec, 17% fewer output tokens, knowledge cutoff updated from January 2025 to March 2026. The honest finding: Artificial Analysis Intelligence Index stayed at 50 — identical to Gemini 3.5 Flash. Faster and cheaper, not smarter. Migrate from 3.5 Flash: yes. Compare against GPT-5.6 Luna ($0.20/M) before migrating from anything else. Full review →
Qwen3.8-Max fact-check: previewed July 19 — not an August launch. No published benchmark table. No confirmed per-token API price ($2/$6/M circulating on X is Grok 4.5's price, not confirmed for Qwen). No open-weight date or license. Alibaba's claim: "second only to Fable 5" — unverified. Test via Token Plan at ~10% of eventual standard pricing. Full fact-check →
August Deadline Calendar — 26 Days Left
✓ Done — August 5: Grok Voice Think Fast 2.0 migration complete. Pin grok-voice-think-fast-1.0 to stay on v1.0.
26 days — August 31: Kimi kimi-k2.5 and moonshot-v1 series sunset. Migrate to kimi-k3 or kimi-for-coding. Migration guide →
26 days — August 31: Claude Sonnet 5 intro pricing ends. $2/M → $3/M (+50%). New tokenizer adds 10-35% more tokens. Front-load batch jobs now. Action plan →
80 days — October 24: deepseek-chat and deepseek-reasoner deprecated. Migrate to deepseek-v4-flash now. V4 Flash 0731 review →