Note: This is an English adaptation of a Chinese forum post discussing the same-night launches of DeepSeek V4 Pro 0813 and Grok 4.6 on August 12, 2026.
Key points
- DeepSeek V4 Pro 0813 and SpaceXAI Grok 4.6 launched within 2 hours of each other on the night of August 12, 2026 (Beijing time), closing out an August-long wave of AI coding backend upgrades.
- DeepSeek V4 Pro 0813 is not a new model: total parameters (1.6T MoE) and 1M context match the April preview. The GA release's gains are all post-training (GA stability + multi-turn Agent RL).
- Grok 4.6 reuses the 1.5T V9 architecture from July; 4.6 is the result of longer SFT, regenerated Grok 4.5 training trajectories, and domain RL. Experts expanded from 128 to 256 while per-token activation stays at 8–16.
- DeepSeek's multimodal "full-strength" version, previewed in April, still had no launch date as of August 13.
- Real-world impact of Grok 4.6's 200K+ price tier needs more deployment data; no public benchmarks on quality beyond 200K.
- DeepSeek's pricing is pre-increase; the size of the future increase and the length of the zero-cost switch window are unknown.
- Neither model provides independent third-party latency/throughput data.
- https://api-docs.deepseek.com/quickstart/pricing
- https://weibo.com/5424490006/5331252565247196
- https://k.sina.com.cn/article_7879923925_1d5ae18d506801mg5e.html
- https://www.cnblogs.com/sing1ee/p/22436555
- https://k.sina.com.cn/article_7879996424_1d5af340806801n246.html
- http://x.ai/news/grok-4-6
- https://tech.ifeng.com/c/8vX3XV7yIcU
- https://singularity.kiwi/spacexai-grok-4-6-frontier-cheap-agents-2026
- http://agentbreaking.com/blog/grok-4-6-deep-dive-benchmarks
Benchmark comparison
| Dimension | DeepSeek V4 Pro 0813 | Grok 4.6 | Claude Fable 5 Max | |---|---|---|---| | Total / activated params | 1.6T MoE / 49B | 1.5T MoE / 8–16 activated | undisclosed | | Context / max output | 1M / 384K | 256K (stable ~128K in testing) / 500K (if benchmark supports) | 200K | | AA Intelligence Index | — | 61 | 62 | | DeepSWE v1.1 | — | 65.9% | 70% | | Terminal-Bench v3.0 | — | 26% | 34.1% | | DeepSWE (preview → GA) | 12.8 → 62.7 | — | — | | Terminal Bench 2.1 (preview → GA) | 72.1 → 87.9 | — | — | | Cybergym (preview → GA) | 52.7 → 83.3 | — | — | | Price ($/MTok in / out) | 0.435 / 0.87 | 2 / 6 | 5 / 25 | | Platforms | API (OpenAI + Anthropic + Responses formats) / Chat / Huawei Ascend | Cursor / Grok Build / OpenRouter / Vercel / Cloudflare | Anthropic API / Bedrock / Vertex |
Same-night product strategy
Within the August upgrade curve of AI coding backends (routers on Aug 6, harnesses on Aug 9–11), these two launches are the "model layer" closing moves: Grok 4.6 defends first-tier status (AA Index 61, matching GPT-5.6 Sol, one point behind Fable 5 Max at half the price), while DeepSeek converts its preview to GA at 1/60 of Fable 5's price. Cursor offered 2× usage for Grok 4.6 in week one; DeepSeek let existing users switch with zero code changes before an announced price increase.
Why DeepSWE matters now
DeepSWE v1.1 tests multi-file modification across a full codebase—closer to real developer workflows than single-file HumanEval. DeepSeek's jump from 12.8 to 62.7 crosses the "commercially usable" threshold; the gap to Fable 5 is estimated at only 5–7pp against a 60× price difference, marking a price-performance inflection point.
Long-horizon agent cost
On the AA-Briefcase long-horizon benchmark, Grok 4.6 used ~53 turns and ~0.5B input tokens versus ~103 turns and ~2B tokens for Claude Opus 5 (max)—half the turns, a quarter the tokens. However, Grok 4.6 has a long-context pricing trap: past 200K context, input price jumps from $2 to $4 and output from $6 to $12, applied to the entire request. Artificial Analysis places Grok 4.6 at $0.84/task on the cost-efficiency frontier, still behind GPT-5.6 Luna / GLM-5.2 / Muse Spark 1.2.
Winners and caveats
1. Price-performance: DeepSeek V4 Pro 0813 wins decisively (60× gap). 2. Multimodality: Grok 4.6 wins (text, image, audio, video, Grok Build integration); DeepSeek still has no vision—the full multimodal version has no release date. 3. Default backend coverage: DeepSeek benefits (triple API format support means most harnesses connected automatically). 4. Long-horizon efficiency: Grok 4.6 wins. 5. Long-context cost: Grok does not necessarily win (price jump above 200K).