THURSDAY JULY 10 — TWO LAUNCHES, ONE DAY
● GPT-5.6 Sol/Terra/Luna: OpenAI. Sol at $5/$30/M (88.8% Terminal-Bench 2.1). Terra at $2.50/$15/M. Luna at $1/$6/M.
● Grok 4.5: SpaceXAI. 1.5T V9 model, Cursor training data, reinforcement learning from SpaceX/Tesla environments. Vendor claim: near-Opus performance. No independent benchmark published yet.
● What we do know about Grok 4.5: Monthly release cadence confirmed. V9 at 1.5T parameters (3x v8-small). Cursor supplemental training. RL still running at SpaceX and Tesla.
● What we do not know: Grok 4.5 pricing. Whether the Opus performance claim holds on independent benchmarks. API availability timeline on Amazon Bedrock.
● The day to watch: Independent LMSYS Arena voting and Artificial Analysis benchmarks will begin accumulating as soon as both models are publicly accessible. The community verdict on Thursday will be the first real data.
Why Thursday July 10 Is a Significant Date
The last time two major AI companies released frontier-tier models on the same day was early 2024 — a period when the pace of releases was accelerating rapidly but the models were not yet at the capability level that triggers government review. Thursday July 10 is different in kind: two models that have each been described as the most capable coding and reasoning systems their respective companies have built, releasing simultaneously to the public. The community will benchmark them against each other immediately. Arena votes, third-party API latency tests, and developer first impressions will all arrive within hours.
Elon Musk's xAI announced that Grok 4.5 would become publicly available at the same time as GPT-5.6's broad rollout. The timing is deliberate. xAI is ensuring that Grok 4.5 enters public consciousness at the moment when developer attention is at its highest — the day GPT-5.6 launches. It is also a signal: xAI is confident enough in Grok 4.5's performance relative to GPT-5.6 to put them in direct public comparison from day one rather than launching separately and being evaluated against each other later.
What Grok 4.5 Brings — Confirmed vs Claimed
Confirmed: V9 foundation model at 1.5 trillion parameters — approximately three times the size of the v8-small model currently serving production on X. Cursor developer workflow data added in supplemental post-pre-training (an xAI engineer noted this is "not quite as good" as having it in initial pre-training — a fuller 2T run with Cursor from the start is underway separately). Reinforcement learning from human feedback running in SpaceX and Tesla engineering environments. Private beta at both companies since June 28.
Claimed but unverified: Elon Musk said early evaluations showed performance "close to, perhaps exceeding Claude Opus." Claude Opus 4.8 scores 69.2% on SWE-bench Pro and leads the Artificial Analysis Intelligence Index at 61.4. No independent benchmark has been run on Grok 4.5. No system card has been published. No third-party evaluation from LMSYS Arena, Artificial Analysis, or Scale SEAL has appeared. Thursday is when independent evaluation begins. Treat the vendor performance claim as directional until the benchmarks land.
The Comparison Matrix — What Thursday's Launches Cover
| Capability |
GPT-5.6 Sol |
Grok 4.5 (claimed) |
Claude Sonnet 5 |
| Terminal-Bench 2.1 |
88.8% (91.9% Ultra) |
TBD Thursday |
80.4% |
| SWE-bench Pro |
Not reported |
TBD Thursday |
63.2% |
| Live X firehose data |
No — Bing search |
Yes — exclusive |
No — web search |
| Context window |
1.05M tokens |
TBD |
1M tokens |
| API price (input/output /1M) |
$5/$30 (Sol) |
TBD Thursday |
$2/$10 intro |
| Government review required |
Yes — 13 days |
Unknown |
No (Sonnet 5) |
What to Watch Thursday — The Key Questions
Grok 4.5 SWE-bench and Terminal-Bench scores: The vendor claim of near-Opus performance implies SWE-bench Pro scores above 60% and Terminal-Bench well above 70%. If independent evaluators confirm this, the three-way comparison between GPT-5.6 Sol, Grok 4.5, and Claude Sonnet 5 resets the competitive landscape entirely. If independent scores fall below the claim, the benchmark gap to Claude Fable 5 (80.3% SWE-bench Pro) remains significant.
Grok 4.5 pricing: If Grok 4.5 is priced near GPT-5.6 Sol ($5/$30/M), it will face direct comparison on price and performance. If it is priced near Grok 4.3's current $1.25/$2.50/M on Bedrock, it becomes the strongest value proposition in the market assuming benchmark parity with Sol. The pricing announcement Thursday morning is the immediate signal.
LMSYS Arena voting: Arena Elo is the most reliable user-preference benchmark because it is based on blind A/B voting at scale. Grok 4.3 already leads at ~1,493 Arena Elo. If Grok 4.5 launches Thursday and Arena voting begins immediately, the first 24-48 hours of Arena data will be the most watched benchmark result of the week.
Source: MLQ News (GPT-5.6 confirmed Thursday), xAI announcements · Related: Grok 4.5 private beta: full analysis → · GPT-5.6 Sol/Terra/Luna pricing and benchmarks → · Grok 4 vs Claude Sonnet 5 vs GPT-5.5 benchmarks →