The AI tools that
actually ship products

Winner

9.7

Claude Opus 5

VS

8.3

Muse Spark 1.1

Claude Opus 5 wins overall, but not because 61 beats 51. It wins because it also reaches 51 — at low effort, for $0.36 per task against Muse Spark 1.1's $0.26 — while keeping four higher rungs that Meta's model cannot reach at any price, and because it can be used outside the United States, which Muse Spark 1.1 currently cannot. The concession is real: at equal measured intelligence Muse Spark 1.1 is about 28 percent cheaper per task, starts responding roughly 48 times sooner and sustains about 2.4 times the throughput. A fixed, validated, US-based workload at index 51 should run on Muse Spark 1.1 and would be overpaying on Opus 5. A workload that might need more than 51, or that serves users outside the US, has exactly one option of the two.

9.7

Claude Opus 5

VS

8.5

GLM-5.2

No overall winner, because the measurements refuse to produce one — and the tie is not where the headline suggests. On Artificial Analysis Intelligence Index v4.1, Claude Opus 5 scores 61 at max effort and GLM-5.2 scores 51 at max effort, a ten-point gap. But the two models meet exactly once, at 51: Opus 5 reaches it at low effort, the bottom of a five-rung ladder, for $0.36 per task, while GLM-5.2 reaches it at max effort, which is both its default and its ceiling, for $0.32 per task. GLM-5.2 therefore wins the tie by about 11 percent and Anthropic's cursor never descends low enough to take that point back. What Opus 5 has instead is headroom: four further rungs ending at 61, at 5.6 times the tie price, against a wall on the GLM-5.2 side that no budget moves. GLM-5.2 also wins on token price ($1.40 and $4.40 per million against $5 and $25), on blended price ($0.90 against $3.85 per million), on throughput and latency, and on 753B parameters of weights published under an unmodified MIT license we verified by reading the file. The decision is a capability threshold, not a budget: run your hardest work against GLM-5.2 at max effort, and if it completes, the ten points are money with no job to do.

Winner

9.7

Claude Opus 5

VS

8.6

GPT-5.6 Luna

Claude Opus 5 wins on capability, headroom, long-prompt cost predictability and time to first token; GPT-5.6 Luna wins decisively on price and ties the one score they share. Both reach 51 on Artificial Analysis Intelligence Index v4.1, but Luna gets there for $0.28 per task against Opus 5's $0.36 — roughly 22 percent cheaper, both measured July 28, 2026, before OpenAI's July 30 price cut — and that 51 is Luna's absolute ceiling while it is Opus 5's floor, with 56, 59, 60 and 61 stacked above it. Since OpenAI cut the GPT-5.6 rates by a factor of five on July 30, 2026, Luna's list price sits 25 times below Opus 5's on input and about 20.8 times below on output, so the price gap is far wider than the July 28 cost-per-task figures suggest. Luna also reprices an entire request at double input and 1.5 times output above 272,000 prompt tokens, where Opus 5 bills its full 1M-token window flat. Pick Luna if the workload genuinely tops out at 51 and prompts stay short; pick Opus 5 if the ceiling is variable, prompts run long, or first-token latency is the felt experience.